System Integration to Intelligent Control: State of the Art and Future Trends of Electric Vehicle Regenerative Braking Systems

Abstract

With the rapid development of the electric vehicle (EV) industry, the regenerative braking system (RBS) has become a pivotal technology for enhancing overall vehicle energy efficiency and safety. This article systematically reviews recent research advances, spanning macro-architecture, drive and energy-storage hardware, control strategies, and evaluation frameworks. It focuses on comparing the mechanisms and performance of six categories of intelligent control algorithms—fuzzy logic, neural networks, model predictive control, sliding-mode control, adaptive control, and learning-based algorithms—and, leveraging the structural advantages of four-wheel independent drive (4WID) electric vehicles, quantitatively analyzes improvements in energy-recovery efficiency and coordinated vehicle-dynamics control. The review further discusses how high-power-density motors, hybrid energy storage, brake-by-wire systems, and vehicle-road cooperation are pushing the upper limits of RBS performance, while revealing current technical bottlenecks in high-power recovery at low speeds, battery thermal safety, high-dimensional real-time optimization, and unified evaluation standards. A closed-loop evolutionary roadmap is proposed, consisting of the following stages: system integration, intelligent control, scenario prediction, hardware upgrading, and standard evaluation. This roadmap emphasizes the central roles of deep reinforcement learning, hierarchical model predictive control (MPC), and predictive energy management in the development of next-generation RBS. This review provides a comprehensive and forward-looking reference framework, aiming to accelerate the deployment of efficient, safe, and intelligent regenerative braking technologies.

1. Introduction

The unprecedented growth in the transport industry by the extremely rapid increase in the adoption of electric vehicles (EVs) is pulling the transport industry into the era of intelligence and electrification, and the regenerative braking system (RBS) is one of the most noticeable technologies with extreme scientific and engineer merits [1]. When regenerative braking converts a vehicle’s kinetic energy into electrical energy at decelerations and stores the current in the battery pack, overall energy utilization and driving range are greatly improved [2]. This pursuit of extended range is further advanced through various energy optimization strategies, including real-time metadata-driven routing to minimize consumption [3] and sophisticated online energy management for hybrid storage systems [4]. Although the first hybrid and electric vehicles decades ago utilized RBSs, there still exist complications in optimizing the efficiency in the various operating conditions by the coordination with mechanical braking approaches and in conquering battery- and motor-crept limiters [5]. Spurred on by the automobile industry’s unrelentless focus on conservations of energies and emissions reductions, the research papers on the energy-recovery systems have grown exponentially in the past two decades and illustrate the extreme enthusiasm of both the research communities and the industry for regenerative braking [6].

At present, research hotspots center on two fronts: the application of four-wheel independent-drive architectures and the development of intelligent control algorithms [6]. On the one hand, distributed four-wheel-drive EVs (i.e., each wheel propelled by an independent motor) have become a research focus because of their advantages in braking energy recovery and stability control [7]. Compared with conventional front- or rear-drive vehicles, all-wheel-drive EVs can harvest significantly more energy during braking; related studies report an increase of approximately 23–31% [8]. This improvement arises from the vehicle’s ability to flexibly allocate braking force between the front and rear axles according to load transfer while coordinating electric and mechanical braking on each axle to maximize energy recovery [9]. Meanwhile, emerging brake-by-wire (a system that replaces traditional mechanical and hydraulic connections with electronic sensors and actuators, allowing for precise computer-controlled braking) and in-wheel motor technologies make it possible to control the braking force of each wheel independently, simultaneously boosting recovery efficiency and greatly enhancing braking safety and vehicle stability; under an all-wheel-drive layout, both safety performance and energy-recovery effectiveness surpass those of traditional single-axle configurations [10].

In contrast, control strategies have experienced a discernible transition from classical towards intelligent methods [11]. In the past years, sophisticated predictive algorithms like model predictive control (MPC) and machine-learning strategies like reinforcement learning have been research foci in RBS control and are increasingly replacing the once universal usage of PID control, making way for more optimized brake-energy management [6]. E.g., MPC maximizes the allocation in regenerative braking force by forecasting future braking requirement based on real-time traffic and vehicle conditions, enhancing energy-recovery efficacy by about 15% in urban cycles [12]. Other research has introduced two-stage nonlinear MPC algorithms that dynamically allocate regenerative and mechanical braking powers and demonstrate over a 24% increase in driving range [10]. Furthermore, reinforcement-learning (RL) algorithms are now being incorporated into multi-objective RBS control where an RL agent learns the optimum strategies by exploring the environment in order to integrate objectives such as energy recuperation and braking safety [13]. The most up-to-date findings indicate RL-based RBS-control strategies drastically dominate conventional optimization strategies in both energy-recovery efficacy and battery life and reach performance close to the DP optimum [13]. Simultaneous utilization of such intelligent-control strategies and independent-drive architectures is driving the development in EV regenerative braking towards more efficiency and more intelligence.

Although several reviews on regenerative braking systems (RBSs) have been published (as shown in

Table 1. Comparison of Existing Review/Survey Papers on RBSs and Intelligent Control.

Reference	Focus	Gap	Novelty/Contribution
Szumska, E. M. (2025) [14].	General EV RBS design, efficiency issues, conventional control	Does not explicitly address 4WID integration or advanced AI algorithms	Provides a comprehensive synthesis of EV RBS design and efficiency challenges, highlighting key structural and control issues.
Vodovozov, V. et al. (2021) [15].	Energy allocation, braking force distribution, hardware limits	Limited discussion on intelligent control and 4WID synergies	Contributes detailed insights into braking energy allocation and system-level management approaches in EVs.
Hang, P. et al. (2021) [16].	Chassis architecture and control perspectives	Does not comprehensively link RBS with intelligent control strategies	Explores 4WID/4WIS vehicle configurations and their control implications for autonomous driving.
Saiteja, P. et al. (2022) [6].	Calibration, architecture, system-level challenges	No 4WID-specific integration, limited discussion of AI/learning control	Highlights critical aspects of system architecture, calibration methods, and implementation challenges in RBS.
Chen, Z. et al. (2025) [17].	BBW hardware and prospects	Touches on RBS but not integrated with 4WID and AI-based control	Maps the state of the art in brake-by-wire technology for new energy vehicles, setting foundation for integration with RBS.
Hamada & Orhan (2022) [18]	Energy-storage devices (flywheels, supercapacitors, hybrid storage)	Lacks in-depth application on 4WID EVs	Discusses the various types of energy-storage devices, providing foundational insights for 4WID integration.
Our review	Systematic comparison of six intelligent control methods (fuzzy, NN, MPC, SMC, adaptive, learning-based) + hardware advances + evaluation framework	——	First integrative review explicitly connecting 4WID with intelligent RBS control, proposes closed-loop roadmap with DRL, hierarchical MPC, predictive energy management

), they remain limited in scope. Most prior works emphasize general RBS design and efficiency issues or focus on braking energy management and allocation strategies, while others highlight 4WID/4WIS configurations or hardware-oriented technologies such as brake-by-wire. Some reviews largely address traditional architectures and calibration issues without systematic coverage of intelligent control methods, whereas others concentrate on specific case studies such as MPC-based fault-tolerant control rather than providing a comprehensive survey.

As shown in Figure 1, the rapid growth of the electric vehicle (EV) industry has underscored the importance of regenerative braking systems (RBSs) in improving energy efficiency and driving range. Despite the progress, previous reviews have not fully captured the recent advancements in intelligent control algorithms and four-wheel independent-drive (4WID) architectures. This review aims to fill this gap by systematically analyzing the latest developments in these areas, focusing on six categories of intelligent control algorithms and their application in 4WID RBS for EVs. This paper provides three unique contributions: (1) a comprehensive system-integration approach that connects RBS architecture, control strategies, and vehicle-level coordination; (2) a comparative analysis of intelligent control methods, including fuzzy logic, neural networks, model predictive control, and reinforcement learning; and (3) a quantitative evaluation of 4WID advantages in energy recovery and vehicle stability compared to traditional drive architectures. By synthesizing these advancements, this review offers valuable insights into the state-of-the-art and future prospects of intelligent RBS technologies for EVs.

The structure of this review is as follows: Section 2 outlines the literature method, following the PRISMA guidelines for systematic review. Section 3 focuses on hardware advancements, including RBS architecture and energy storage systems, while Section 4 delves into the integration of four-wheel independent-drive (4WID) with regenerative braking systems. Section 5 provides an overview of intelligent control methods, such as fuzzy logic, neural networks, model predictive control, and reinforcement learning. Section 6 presents a comparison of these control strategies, particularly in their application to 4WID systems. Section 7 discusses the challenges and trends in this field, and Section 8 concludes with insights into the future of intelligent RBS technologies for electric vehicles.

2. Literature Selection and Data Extraction Methodology

To ensure the credibility and rigor of the literature selection process, a systematic review methodology was employed following the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. As shown in Figure 2, below is a detailed account of the steps involved in the selection of studies for inclusion in this review.

• Identification of Studies.

The literature search was conducted using academic databases such as IEEE Xplore, ScienceDirect, Scopus, and Web of Science. A combination of search terms like “regenerative braking systems,” “electric vehicles,” “intelligent control algorithms,” and “energy recovery efficiency” was used to capture a wide range of articles related to regenerative braking in electric vehicles. In total, 1382 records were initially identified through database searches, with an additional 15 articles found through other sources (e.g., citations in relevant studies).

• Screening of Articles.

After removing duplicate entries, 1315 unique records remained. The next step involved screening the titles and abstracts of these articles to determine their relevance to the research question. A total of 1160 articles were excluded at this stage because they did not meet the inclusion criteria, such as being unrelated to regenerative braking or focusing on non-electric vehicle applications.

• Eligibility Assessment.

Following the initial screening, 300 full-text articles were retrieved and assessed for eligibility. These articles were evaluated based on the following criteria:

Inclusion Criteria: Peer-reviewed studies discussing regenerative braking systems for electric vehicles, control strategies (such as fuzzy logic, neural networks, model predictive control), and energy recovery performance.

Exclusion Criteria: Articles not relevant to regenerative braking in electric vehicles, articles with significant methodological flaws, and those not peer-reviewed.

At this stage, 151 articles were excluded for reasons such as not directly addressing regenerative braking (n = 73); not focusing on electric vehicles (n = 38); being non-peer-reviewed sources like conference abstracts (n = 31); having methodological issues (e.g., lack of experimental validation or unclear system descriptions, n = 9).

• Final Inclusion.

After the eligibility assessment, 152 articles remained for the qualitative synthesis, providing comprehensive coverage of the field. These studies were analyzed to extract data on control strategies, energy efficiency improvements, and the integration of various technologies such as high-power density motors and hybrid energy storage systems. Out of these, 62 articles were included in the quantitative analysis due to their provision of robust data suitable for meta-analysis.

• Data Extraction and Synthesis.

Data from the 152 articles were carefully extracted, including information on types of regenerative braking systems used; control strategies (fuzzy logic, neural networks, model predictive control, Sliding Mode Control Method, Adaptive Control Method); performance metrics (energy recovery efficiency, braking stability, and real-time optimization); technological advancements (e.g., high-power density motors, hybrid energy storage, brake-by-wire systems); challenges such as high-power recovery at low speeds and battery thermal safety.

The studies were synthesized to identify emerging trends, research gaps, and potential areas for future innovation in regenerative braking systems for electric vehicles. The meta-analysis focused on quantifying improvements in energy recovery efficiency and evaluating the effectiveness of different control strategies.

3. RBS and Power Storage Technologies for EVs

3.1. Control System Architecture of RBS in EVs

Electric vehicles (EVs) need a high-efficiency brake controller in order to obtain the maximum braking performance and safety. The function of the Regenerative Braking System (RBS) is regenerative-energy efficiency enhancement, braking stability promotion, and performance optimization. As depicted in Figure 3, we demonstrate the common control architecture of the RBS in EVs [6]. The architecture consists of various subsystems and control units whose coordination optimizes the energy recovery and the safety control by fusing the multi-source signals from the sensors.

First, real-time inputs such as vehicle speed, steering angle, brake-pedal position (BPP), motor and battery states, state of charge (SOC), temperature, and voltage are input into the driver-behavior recognition and vehicle dynamics prediction unit, which simultaneously evaluates vehicle state, wheel–speed slip ratio, and driver braking requirement. The results are then outputted to the brake-control unit and the regenerative-brake controller: the former calculates the necessary mechanical-braking torque, whereas the latter predicts the optimum regenerative-braking force based on conditions of the motor and battery. Within this scheme, state-of-the-art control strategies such as neural networks, model-predictive control, and sliding-mode control are incorporated in order to actively distribute hydraulic and regenerative braking forces in a way maximizing energy recuperation and braking stability under various operating conditions. As illustrated in Figure 3, the RBS forms a closed loop between the decision layer and the execution layer: the brake-control unit outputs the optimal braking-force distribution to the motor-control unit and the hydraulic-actuation unit, thereby realizing the cooperative operation of regenerative and hydraulic braking [19]. During emergency braking, hydraulic and regenerative braking act simultaneously; the motor, operating in generator mode, converts the wheels’ rotational kinetic energy into electrical energy that is stored in the battery. The RBS also performs energy and power analyses to determine how to distribute the stored energy to various vehicle systems with maximum efficiency. Consequently, it is essential to gain a comprehensive understanding of the RBS, including all major control modules involved in processing input information. Such a critical analysis can be carried out using the entire RBS architecture, which helps researchers grasp the process flow of input data and identify the key modules involved in decision-making [20,21].

3.2. Comparison of Common Motor Types

In the regenerative braking (RB) process in the EV, the traction motor acts as a generator and converts a certain amount of the vehicle’s kinetic energy into electrical energy and thus lowers the net amount of consumed energy. Meanwhile, it has the potential for a rapid torque response for braking and raises the vehicle dynamic performance. The most popular types applied in the EV RB are the Brushed Direct-Current (DC) motor, the Induction motor (IM), the Permanent-Magnet Synchronous Motor (PMSM), and the Switched Reluctance Motor (SRM). Their operating principles, performance characteristics, and applications are different as follows:

• Brushed Direct-Current motor.

A Brushed DC has a simple structure in which the DC is provided by the carbon brushes and the commutator to the armature windings in generating the torque. The motor has simple control and favorable starting characteristics but low power density, average efficiency, and brush wear. With regenerative braking, feedback of the energy is possible by the four-quadrant driver, but the low conversion efficiency and the correspondingly high maintenance cost render the motor hardly implemented in current-generation EVs [22].

• Induction motor.

An IM has essentially three-phase stator windings and a squirrel-cage rotor; the rotating magnetic field induces rotor currents in order to produce the torque. IMs are rugged, do not require the rare-earth magnets, and are very economical. Their efficiency maximum is typically 85–90%, although at light-load the power factor is low. They have a wide speed range and good hot-temperature performance, but the rotor currents contribute additional losses and heat. Regeneration demands sophisticated inverter control, now mature; efficiency and dynamic response at the low speed by IM are slightly worse in comparison with permanent-magnet machines [23].

• Permanent-Magnet Synchronous Motor.

A PMSM has rotor-embedded permanent magnets; stator currents in the three-phase winding generate a rotary field synchronously rotating the rotor. Because of extremely high efficiency (typically above 95%) and torque density, PMSMs have dominated the traction motor in the majority of EVs. PMSMs possess a wide constant-torque and constant-power speed range, can generate excellent electromagnetic torque at low speed, and possess rapid response. Vector-control algorithms (advanced motor control methods that independently manage the magnetic flux and torque-producing components of motor current for superior performance) are required for control, which is relatively more complex compared with BDCs or IMs. PMSMs are expensive (rare-earth material) and require high-performance inverters, but many mass-production vehicles—such as the Tesla Model 3 and most Chinese new-energy vehicles—adopt PMSMs in order to expand range and power performance [24]. In addition to Tesla, other leading OEMs such as BYD and NIO also widely adopt PMSMs in their electric vehicle platforms (e.g., BYD Han, NIO ET7) due to their high efficiency and power density. Toyota, on the other hand, has utilized induction motors in some models like the earlier Tesla Roadster and continues to explore hybrid configurations that integrate regenerative braking with hybrid energy management systems [25,26].

• Switched Reluctance Motor.

The SRM has simple geometry with no permanent magnets or stator windings. When stator coils are excited, the rotor is attracted to the position of minimum magnetic reluctance. Advantages are strength in structure, low cost, and wide constant-power region. Because the rotor contains no windings, the generated heat is localized in the stator and susceptible to cooling as well as operation in high-temperature and high-vibration applications. High-efficiency SRMs in current designs have reached near-PMSM performance levels in power density and efficiency. However, simple nonlinearities cause ripples in the torque, vibrations, and noises; efficiency remains moderate, poorer compared to PMSMs, and SRMs require special power converters and sophisticated control approaches. In general terms, SRMs are appropriate in applications where strength and cost are paramount [27].

Table 2. Performance comparison of commonly used EV motors [1,28].

	Brushed DC	IM	PMSM	SRM
Motor structure diagram
Power density	2.5	3.5	5.0	3.5
Efficiency	2.5	3.5	5.0	3.5
Weight	2.0	4.0	4.5	5.0
Controllability	5.0	5.0	5.0	3.0
Reliability	3.0	5.0	5.0	4.5
Technology maturity	5.0	5.0	5.0	4.0
Cost	4.0	5.0	3.0	4.0
Total	24.0	31.0	32.5	28.0

presents a comparative analysis of the motors most commonly used in EVs—BDC, IM, PMSM, and SRM. The comparison considers efficiency, weight, and cost. Each motor is rated on a scale from 1 to 5, where 5 indicates the best performance.

The comparison above shows that PMSMs have become the mainstream motor for today’s EVs thanks to their high efficiency and power density; IMs, although reliable, are slightly less efficient; SRMs feature a simple structure, but control complexity and torque characteristics limit their widespread adoption; and BDCs have been largely replaced by brushless machines in new-energy vehicles because of efficiency and maintenance concerns.

3.3. Energy-Storage Units

Since the RBS needs to reserve the recovered braking energy for later utilization, the selection of an adequate energy-storage unit is essential. Popular technologies are the lithium-ion battery (LIB), the supercapacitor (SC), the flywheel energy-storage system (FESS), and the hybrid energy-storage system (HESS). Comparisons between the principal technical indices of the above four choices are given in

Table 3. Comparison of key technical indicators for energy-storage units [29,30,31].

Indicator	Lithium-Ion Battery (LIB)	Supercapacitor (SC)	Flywheel Energy-Storage System (FESS)	Hybrid Energy-Storage System (HESS)
Energy density (Wh kg−1)	150–250	5–10	20–80	Depends on combination (between individual units)
Power density (W kg−1)	250–700	500–10,000	500–1500	Comprehensive result (composite)
Cycle life (cycles)	≈2000–3000	≥100,000	≈20,000	Determined by the shortest-lifespan component
Storage efficiency (%)	≈90–95	≥95	≈90–95	High; strategy-dependent

.

LIB stores chemical energy by intercalation and de-intercalation of lithium ions within the electrode materials and is noted by superior energy density, good storage efficiency, and a mature management system, which makes it a prospective technology for long-duration energy storage [32]. Its power density is low; aggressive C-rate charging by repeated regenerative braking has the potential to accelerate degradation or lithium plating; and the product cost by capacity is still relatively high. Research work is therefore focused on optimization of the electrode material, development in the electrolyte, or hybridization of the LIB with other storage devices in attempts to extend the high-rate capability and service life [33].

The SC is electrostatically kept in store by the effect of the electric-double-layer with very long cycle life and high power density and therefore responds quickly to the pulsed high powers from regenerative braking by the RBS. Its energy density is nonetheless very low, the self-discharge is heavy, and the cost is costly; the voltage control is moreover cumbersome. A standalone SC consequently rarely meets the large-scale requirement for store in an RBS and is employed almost exclusively as the secondary device in parallel with a battery [29]. The FESS provides kinetic energy from a flywheel operating at extremely high speed and has the feature of rapid response, potential for high power density, high efficiency, and long cycle life with non-chemical pollution. Its energy density is only moderate; device mass and device cost are significantly large; mechanical loss, complication in control circuitry, and harsh safety conditions restrict common applications in passenger EVs. FESSs are applied almost exclusively in heavy-duty vehicles or rail transportation [34].

An HESS combines a number of devices—typically a battery and an SC or FESS—in attempting to reconcile the discrepancies in the power and the energy densities and the optimum efficiency in the overall system. At regenerative braking, a device well suited for power-dominated applications (SC or FESS) accumulates the ensuing high-power energy preferentially, with the battery supplying the continuing energy and hence mitigating the adverse effect on the battery from the ensuing high-rate cycling. Zhao et al. [35] in Figure 4 conducted a comprehensive analysis on the LIB–SC HESS and designed an optimum power distribution scheme whose promise was great in simulation in decreasing the consumed energy by about 8.9% and improving the state performance significantly in the battery and the SC. An HESS has the potential benefit but is more complex and adds more power-electronic equipment and sophisticated control strategies, and applications are still at prototype and small-scale-demo level.

Beyond capacity and power, the impact of regenerative braking on battery degradation and safety is a critical system-level concern. The high-current, pulsed charging characteristic of RBSs can accelerate battery aging mechanisms such as lithium plating at the anode (especially at low temperatures or high SOC) and growth of the solid electrolyte interphase (SEI) layer, leading to irreversible capacity fade and increased internal resistance [2,36]. Consequently, the Battery Management System (BMS) must impose stringent limits on regenerative power to prevent overcharging and mitigate degradation, often curtailing the theoretical recovery potential. Future RBS designs require integration with battery health (SOH) and thermal models to enable a lifecycle management approach that optimally balances immediate energy recovery against long-term battery longevity and safety.

4. Analysis of the RBS and Four-Wheel Independent Drive Architecture

4.1. RBS Characteristics of Different Drive Configurations

The RBS, the main technology for making EVs more energetically efficient, was first applied in the original HEVs. In the HEV, the traction motor operates in generating mode during braking and converts a fraction of the kinetic energy into storable electrical energy in the battery, expanding driving range and reducing mechanical-brake wear. On the original HEVs, however, low-motor power and continued operation of the internal-combustion engine and conventional drivetrain restricted the performance potential of the regenerative braking. With the development of battery electric vehicle (BEV) technology, traction systems have taken on more electrification, integration, and power density, permitting the full potential of the RBS to be actualized.

As depicted in Figure 5, the EVs can be classified into the following three configurations in consideration of the position and the number of the traction motors: single-axle drive, two-axle drive, and four-wheel independent drive. Such configurations greatly influence RBS energy-recovery efficiency and control strategies [1].

• Single-axle-drive EVs

In a single-axle setup—with front-wheel or rear-wheel drive—the motor is placed on one axle. While braking, the former recovers the energy only; the other one has the sole option of friction braking; hence, some amount of the kinetic energy is lost. Hence, the front-drive models put maximum braking loads on the front wheels by virtue of load transfer; the rears, incapable of regeneration, take up the same in the form of heat. Rear-drive models are constrained by the requirement for front-axle braking in hard braking or downhill driving conditions, even more restricting the potential for recovery [37]. Latest research results indicate single-motor EVs have a braking-energy recovery efficiency of 14–20% under normal urban cycles, 9–12% under highway operating conditions, and still possess braking-bias problems and early ABS kick-in on icy conditions [14].

• Dual-axle-drive EVs

A two-motor architecture (one per axle) involves all four wheels in recuperation and therefore drastically enhances potential recuperation. At light-to-moderate braking, simultaneous generation of regenerative torque from the front and rear motors supplies the ideal front-to-rear distribution and improves vehicle stability, in addition to reducing the employment of friction brakes. The two-motor setup has the side benefit of control redundancy and fault tolerance: if one motor is disabled by overheat or by some SOC limit, the other one can by itself accomplish part of the regenerative task and forestall efficiency and stability from being compromised. Latest measurements show two-motor EVs recuperate 22–28% in normal cycles and up to 32% with optimized control strategies [38]. Compared with single-axle configurations, efficiency is enhanced by roughly 20–30%; however, front–rear coupling and left–right wheel connection inhibit the control freedom and leave potential efficiency and dynamic-control unexploited [39].

• Four-wheel independent drive (4WID) EVs

The four-wheel independent-drive architecture—in which the wheel is driven by a wheel-end or in-wheel motor, typically—is the future for the EV drivetrain. It has inherent qualities for the RBS: all wheels are independent energy-recovery devices, with all-wheel regeneration on braking; independent braking-torque control at the wheel enables varied tire adhesion, load transfer, and road conditions; and the inclusion of higher-order vehicle dynamics (e.g., torque-vectoring control) becomes tractable, balancing energy recuperation with longitudinal and lateral stability [40]. Latest research results indicate that adequate distribution of regenerative torque from four motors yields 30–38% overall recovery efficiency; in worst-case scenarios—such as cornering brakes or split-μ roads—the stability and attitude disturbances are greatly enhanced by displaying unmitigated supremacy over other architectures [13,41]. As a direct outcome, the four-wheel-drive EV has uncontested qualities for the RBS in imposing sizeable whole-vehicle efficiency and safety gains.

4.2. Distinct Advantages of the 4WID Architecture for Regenerative Braking

As EV propulsion systems advance toward highly integrated, distributed layouts, the 4WID Iconfiguration is emerging as particularly well-suited to regenerative braking. Figure 6 distills its principal benefits into three dimensions.

• Expanded control authority afforded by the multi-motor topology

Because each wheel is propelled by its own traction machine, torque can be metered with exceptional precision. In contrast to axle-coupled drivetrains, 4WID reallocates braking force among the wheels in real time, improving energy-recovery efficiency while enhancing fault tolerance: if one motor must be derated, the remaining units can instantly take up the slack, thereby satisfying functional-safety criteria [42,43].

2.

Higher theoretical energy-recovery ceiling

In single- or dual-motor layouts, the kinetic energy associated with undriven wheels is inevitably dissipated through the friction brakes [37]. A four-motor system, by contrast, enlists every wheel in regeneration and—under light-to-moderate deceleration—can rely on purely electric braking [17,44]. Comparative tests show that, relative to a representative two-wheel-drive architecture, 4WID boosts recovery efficiency by roughly 20–35% over the New European Driving Cycle (NEDC), with the advantage most pronounced in urban traffic [8,38].

3.

Adaptive torque-distribution capability

During regeneration, torque is dynamically apportioned according to motor-efficiency maps, battery state of charge, thermal loading, and longitudinal load transfer [45]. Should the battery approach its charge-current limit or any motor overheat, coordinated multi-motor control immediately reduces the burden on the stressed component and reroutes energy along the optimal path [37]. This strategy extends battery life while improving system stability and thermal management [46].

4.3. Composite Braking-System Architectures

To achieve coordinated control of energy recovery and safe braking, the mechatronic composite braking system of an EV can typically be classified and analyzed from three perspectives—braking-architecture mode, pedal-coupling method, and friction-brake actuator structure—as summarized in

Table 4. Simplified Comparison of Mechatronic Composite-Braking Architectures [42,47].

Dimension	Type	Features	Advantages	Limitations
Braking architecture	Parallel	Fixed regenerative and friction braking ratio	Simple, retains conventional layout	Low recovery efficiency; lacks dynamic adjustment
	Series	Regen-first strategy; dynamic friction braking	Higher recovery efficiency; consistent pedal feel	Complex control; limited dynamic stability
Pedal-coupling method	Coupled	Mechanical linkage	Simple, intuitive	Low control freedom
	Decoupled	Brake-by-wire with simulator	Flexible torque allocation, better control	More complex, higher cost
Friction-brake actuator	Hydraulic	Vacuum booster and hydraulic system	Low cost, redundancy	Engine-dependent, not suitable for pure EVs
	EHB	Motor-driven pump + electro-hydraulic valves	Fast response, integrates easily	Moderate cost, hydraulic components needed
	EMB	Direct motor drive, no hydraulics	Ultrafast response, fully by-wire	Still under development, no mechanical backup

.

Viewed across the historical arc of EV brake development, three interlocking trends stand out:

• System architecture has shifted from a parallel to a predominantly series layout.
• Pedal design has moved from direct mechanical coupling to an electronically decoupled interface.
• Actuation hardware has progressed from a conventional hydraulic circuit to brake-by-wire devices—most commonly the electro-hydraulic brake (EHB) and, in research settings, the electro-mechanical brake (EMB) [48].

Collectively, these changes have not only raised the ceiling on energy recuperation but have also delivered palpable gains in response time, pedal refinement, and overall vehicle controllability.

This evolution towards series decoupled braking architectures is not merely academic but is now standard in production vehicles from leading manufacturers. Tesla’s regenerative braking system, integrated with its one-pedal driving feature, is a prominent example of a series-style, regen-first strategy, heavily reliant on its sophisticated brake-by-wire system for seamless blending. Similarly, BYD’s intelligent braking systems in its e-platform 3.0 vehicles (e.g., Dolphin, Seal) utilize an electronic stability program (ESP) with integrated brake-by-wire functionality to achieve high energy recovery rates while maintaining stability. Suppliers like Bosch and Continental provide core components such as the iBooster and MK C1 electro-hydraulic brake systems, which are employed by a wide range of OEMs including Toyota (bZ4X), NIO, and XPeng to enable these advanced regenerative braking functionalities.

• Architectural perspective

Early mild hybrids favoured parallel layouts for their simplicity, locking electric and friction brakes into a fixed ratio that cannot adapt to real-time demand; regeneration therefore dwindles whenever strong deceleration is requested. A series system reverses that priority: regenerative torque is applied first, with friction braking added only as necessary. Coupled with a brake-by-wire supervisor for fine torque allocation, the series topology now underpins most contemporary BEVs and full hybrids [8].

• Pedal interface

A mechanically coupled pedal provides familiar hardware and direct feedback, yet the linkage constrains control authority. Replacing it with a decoupled (by-wire) pedal and feel simulator digitises the driver’s command, allowing the controller to blend regenerative and hydraulic torque while keeping pedal feel uniform. Nearly all modern hybrids and BEVs have adopted this scheme to improve both performance and comfort [49].

• Actuator technology

Conventional hydraulic systems depend on an engine-driven vacuum booster and are giving way to the EHB shown in Figure 7 [50]. In an EHB, an electric motor and electronically controlled valves regulate wheel-circuit pressure independently, yielding rapid response and built-in redundancy—hence their widespread use in production EVs. The EMB pushes the concept further, dispensing with hydraulics altogether for even faster actuation and higher precision. Lacking a mechanical fallback, however, EMBs remain in R & D and validation and are unlikely to reach mass production in the near term [51].

In summary, a composite braking configuration that marries a series architecture with a decoupled pedal and an EHB actuator has become the mainstream technical route for EV brake control. Future research will concentrate on co-optimising energy-recovery efficiency and braking stability, embedding intelligent control algorithms more deeply in supervisory logic, and advancing high-safety, fully electronic braking systems toward commercial readiness.

4.4. Traditional Control Strategies for RBS in EVs

In traditional EV composite braking systems, researchers have identified two primary strategies for distributing braking forces to balance energy recovery with driving safety: fixed-ratio and rule-based (or look-up-table) approaches. The fixed-ratio method, which is straightforward and easy to implement, establishes a constant ratio between regenerative and friction braking to comply with regulatory standards. This method was particularly useful in early front-drive vehicles due to its simplicity, low cost, and reliability. However, it falls short in terms of adaptability, struggling to account for factors such as load transfer, road surface conditions, and fluctuations in motor and battery performance. As a result, energy recovery can be inadequate, and vehicle stability can suffer, especially in dynamic driving scenarios [39,52].

To overcome these drawbacks, rule-based or look-up-table methods have become the norm in production vehicles. These strategies adjust the distribution coefficients based on data derived from real-world testing and calibration, allowing for dynamic responses to variables such as vehicle speed, deceleration, SOC, and road conditions. This adaptability makes the rule-based approach more practical in comparison to the fixed-ratio method. However, these systems still rely heavily on manual calibration and empirical data, limiting their effectiveness in extreme conditions like inclement weather or non-linear tire behaviors that fall outside the calibrated data set. Furthermore, the complexity of these strategies, which involve numerous parameters and intricate logic, introduces substantial calibration challenges and often hinders efforts to achieve global optimization [46,53].

While these approaches work reasonably well in single-axle or dual-motor configurations, they become less effective in 4WID systems. In a 4WID system, each wheel is individually controlled, significantly increasing the dimensionality of the braking control problem. With eight degrees of freedom to manage, the system must balance multiple objectives, including energy recovery, safety, and comfort, while addressing non-linear constraints related to tires, motors, and batteries. Traditional methods struggle to handle the complexity of this input space and the coordination of multiple actuators, often relegating them to baseline strategies that are insufficient for optimal performance [15].

The introduction of 4WID systems amplifies the coupling and non-linearity of braking control, necessitating real-time monitoring of factors like wheel adhesion, speed differences, and motor braking capabilities. Traditional look-up-table methods are no longer sufficient to manage these complexities, pointing to the need for intelligent algorithms capable of self-learning and dynamic optimization. In recent years, techniques such as model predictive control (MPC) and reinforcement learning (RL) have been explored to strike a balance between energy recovery and stability under multi-constraint conditions. This highlights the growing limitations of traditional strategies in high-dimensional distribution and multi-objective coordination, underscoring the importance of developing more intelligent and adaptable approaches [40].

5. Intelligent Control Strategies for RB Energy Recovery in EVs

5.1. Fuzzy Logic Control (FLC) Method

Fuzzy Logic Control (FLC) has emerged as a prominent approach in the management of regenerative braking energy recovery in EVs, owing to its ability to effectively address uncertainty and non-linear dynamics. As illustrated in Figure 8, FLC mimics human decision-making processes by translating input variables—such as speed, BPP, and SOC—into linguistic terms (e.g., “Low,” “Medium,” “High”) using membership functions. The system then applies a fuzzy rule base for inference and defuzzification, generating precise control signals to regulate the distribution of braking forces between the hydraulic and regenerative braking systems. This optimization enhances both energy recovery and safety. Notably, FLC does not rely on exact mathematical models, making it particularly suited for handling multi-objective, multi-constraint, and highly non-linear scenarios, offering substantial advantages in these contexts.

The core of FLC is the application of fuzzy inference systems (FIS), where control output $u(t)$ is derived from a set of fuzzy rules. The output can be calculated by:

$$u(t)=\sum _{i=1}^N{w}_i\cdot f_i(x_1,x_2,\dots ,x_n)$$

(1)

where $u(t)$ is the control output (braking force); $w_i$ is the weight of the fuzzy rule; $f_i(x_1,x_2,\dots ,x_n)$ is the fuzzy rule function representing the relationship between input variables and output. This formula highlights how the fuzzy control system aggregates the influence of each fuzzy rule to determine the final control output. The weights and the fuzzy functions are determined based on expert knowledge or real-time data.

A substantial body of research has validated the efficacy of FLC in enhancing energy recovery processes. Mei et al. [54] integrated sliding mode control with FLC to optimize braking force distribution by estimating the friction coefficient, which contributed to both improved energy recovery and system robustness. In subsequent work, Mei et al. [55] proposed an adaptive fuzzy sliding mode control strategy to further improve system adaptability across varying road conditions. Wen et al. [56] implemented adaptive FLC for single-pedal control, significantly reducing the need for mechanical braking. Additionally, Li et al. [57] utilized an Adaptive Neuro-Fuzzy Inference System (ANFIS) to lower fuel consumption in hybrid vehicles. Tang and Zuo [58] applied ANFIS to predict driver braking intentions and optimize dynamic braking strategies.

Exploring different motor types and energy management schemes, Liu et al. [24] developed a multi-braking-zone FLC specifically designed for PMSM. Karabacak et al. [59], on the other hand, introduced Type-2 FLC (T2FLC) for Brushless DC Motors (BLDC), aiming to address inherent uncertainties. Aswathi et al. [60] and Nian et al. [61] combined fuzzy logic with inverters and Proportional-Integral-Derivative (PID) control to enhance the performance of BLDC motors. Furthermore, Huynh et al. [62] and Favilli et al. [63] proposed fuzzy coordination strategies applicable to both hybrid and fully electric vehicles, while Chen et al. [64] integrated PID with Anti-lock Braking System (ABS) to optimize system coordination. In the realm of HESS, Hamid et al. [65] and Angundjaja et al. [66] employed FLC to mitigate battery fluctuations, while Cai et al. [67] demonstrated fuel savings achieved through the same method. Additionally, Ning et al. [68] applied fuzzy Q-learning techniques to hydraulic hybrid vehicles, significantly improving energy recovery efficiency. Figure 9 further demonstrates the work by Chen et al. [69], who incorporated Fuzzy-TD3 to enhance both energy recovery and system responsiveness in electro-hydraulic composite braking systems.

In coordinated with the sliding mode control method, Liu et al. [70] proposed an optimal slip-ratio FLC to minimize battery losses, while Subramaniyam et al. [71] combined integral sliding mode techniques to reduce slip error. Peng et al. [72] and Zheng et al. [73] optimized torque coordination and braking mode transitions for 4WID vehicles using fuzzy logic algorithms, successfully balancing safety, energy efficiency, and energy recovery.

In conclusion, Fuzzy Logic Control, with its inherent adaptability, robustness, and integration capabilities, has established itself as a key approach in the energy management of regenerative braking systems within electric vehicles.

5.2. Neural Network Control Method

Neural Network (NN)-based control methods have garnered widespread adoption in the regenerative braking energy management systems of EVs and hybrid electric vehicles (HEVs) due to their remarkable adaptability, ability to model non-linear relationships, and capability to learn from complex systems. As depicted in Figure 10, this control strategy typically involves a three-layer neural network architecture (input layer, hidden layer, and output layer) to process input data and generate appropriate output based on the trained model. The data collected from various driver inputs—such as vehicle speed, BPP, and SOC—are processed according to a pre-trained neural network model, which subsequently produces the precise braking torque signals required for the effective operation of different braking systems. This facilitates the maximization of energy recovery while ensuring safe and stable vehicle operation.

The neural network control system utilizes a network with an input layer, hidden layer, and output layer, which processes inputs and outputs accordingly. The classic formula for a feed-forward neural network with one hidden layer is given by:

$$y=f\left(\sum _{i=1}^n{w}_i{x}_i+b\right)$$

(2)

The formula calculates the output y by taking the weighted sum of the inputs $x_i$, where $w_i$ represents the weights of the connections between the neurons, and b is the bias term. The sum of these weighted inputs is then passed through an activation function f(⋅), such as sigmoid or ReLU, which introduces non-linearity to the system. This allows the network to model complex relationships between the inputs and the output, in this case, generating the braking torque needed to maximize energy recovery in regenerative braking systems.

Early works in this field include Gao et al. [74], who introduced a neural network-based control method for SRM, optimizing power and energy distribution to enhance recovery efficiency. In a similar vein, Zhang et al. [75] integrated Particle Swarm Optimization (PSO) and Ant Colony Optimization (ACO) with NN to propose a predictive regenerative braking control strategy that achieves high efficiency under complex driving conditions. He et al. [76] designed an Intelligent Braking System (IBS), which combined single-pedal control with a multi-objective neural network control method, as shown in Figure 11, resulting in a 3.67% improvement in energy efficiency under NEDC conditions.

The integration of neural networks with other control technologies has also attracted considerable attention in recent research. For instance, Won and Li [77] combined NN with dynamic programming for Interior Permanent Magnet (IPM) motors, significantly improving both traction and recovery performance. Cao et al. [78] merged NN with sliding mode control (SMC) to adaptively adjust control gains online, improving the robustness and efficiency of the recovery process. Rezk and Abuzied [79] utilized Artificial Neural Networks (ANNs) to predict recovery current and parking time, substantially enhancing energy recovery efficiency. Additionally, Shijil and Sindhu [80] incorporated NN into single-pedal regenerative braking systems to improve system stability and reduce driver load. Indu and Aswatha Kumar [81] developed an adaptive NN braking system, which optimized energy efficiency in complex scenarios through simulations in MATLAB/Simulink and IPG Carmaker.

Further advancements include the work by Wang et al. [82], who proposed a dual-mode cruise control system based on NN and MPC to optimize hydraulic braking coordination. Xing and Lv [83] employed Long Short-Term Memory Recurrent Neural Networks (LSTM-RNN) for dynamic brake-state estimation, thus improving management accuracy. Wu et al. [84] integrated Backpropagation (BP) neural networks with PID control to enhance the lateral stability control of Electronic Stability Programs (ESP). Cheng and Ye [85] used a combination of Genetic Algorithms (GA) and NN to optimize energy recovery strategies for electric motorcycles, enhancing battery life. Moreover, Gounis and Bassiliades [86] combined NN with sliding-mode control to bolster Autonomous Emergency Braking (AEB) safety. Vodovozov et al. [87] proposed a NN-based Model Reference Control (MRC) method aimed at maximizing energy recovery while avoiding slip.

In recent studies, Ning et al. [88] employed ANFIS and the Salp Swarm Algorithm (SSA) to optimize the control of the EMB single-pedal mode, achieving recovery rates of 21.54% and 25.39% under WLTC and CLTC conditions, respectively. Shetty and Karabasoglu [89] reported a 37% recovery rate in urban driving conditions using ANN control. Li et al. [90] combined Fuzzy Neural Networks (FNN) to optimize the distribution of hydraulic and electric braking forces, resulting in recovery rates of 14.52% and 39.61% under NEDC and FTP-75 conditions, respectively.

In conclusion, neural network-based control, with its robust data processing, real-time adaptive optimization, and integration with advanced algorithms, has emerged as a cornerstone in the regenerative braking energy management systems of EVs and HEVs. By dynamically adjusting braking torque and recovery strategies in response to complex system conditions, NN control significantly enhances both efficiency and system stability. The continued integration of NN with optimization algorithms will undoubtedly propel the future development of regenerative braking technologies.

5.3. Model Predictive Control Method

MPC is a control strategy that forecasts future system states based on a dynamic model and optimizes the control inputs in real time. MPC has become widely recognized for its application in regenerative braking systems for EVs and HEVs due to its capacity for multi-constraint, multi-objective optimization. As illustrated in Figure 12, MPC processes input signals—such as speed, BPP, and SOC—via a predictive algorithm. The output from this model generates the optimal distribution of braking forces between regenerative braking and hydraulic braking, ensuring both energy recovery and system stability, particularly under complex operational conditions.

A considerable body of research has demonstrated the effectiveness of MPC in optimizing regenerative braking systems. Shi and He [50] introduced an EHB control strategy that addresses hydraulic nonlinearity, thereby achieving smooth and precise pressure tracking. Tang et al. [10] utilized MPC in electro-hydraulic composite braking to coordinate braking torques between the front and rear axles, resulting in enhanced energy recovery efficiency. Valladolid and Macas [91] optimized energy recovery by tailoring control strategies to individual driving habits. Wagner et al. [92] incorporated the consideration of auxiliary loads, such as air conditioning, to optimize overall energy distribution. Zhu et al. [93] combined Nonlinear MPC with PSO to improve recovery efficiency in vehicles equipped with in-wheel motors. Furthermore, Caiazzo et al. [94] focused on fleet-level control, achieving significant energy savings through coordinated management.

In the context of hybrid and ABS coordination, Du et al. [95] proposed an MPC-based strategy that effectively balances recovery efficiency and stability. Figure 13 shows how Pan et al. [96] integrated open-loop and closed-loop models to predict the speed of a preceding vehicle, proposing Eco Adaptive Cruise Control (EACC) to enhance speed tracking and reduce energy consumption. Kumar [97] integrated auxiliary units to reduce the dependency on battery power. Zhao et al. [98] optimized four-wheel-drive motor control using MPC to enhance yaw stability, while Yu et al. [99] applied MPC to optimize load sharing between the engine and the battery. Li et al. [100] leveraged nonlinear MPC combined with PSO to optimize energy recovery efficiency in hybrid buses.

Further advancements in MPC applications include the work by Huang and Wang [101], who utilized nonlinear MPC to optimize braking force distribution between the front and rear wheels, preventing lock-up on low-adhesion surfaces. Nadeau et al. [102] proposed a dual-regeneration and hydraulic braking coordination method. Wei et al. [12] demonstrated a significant reduction in battery energy consumption (by 12–17%) through the implementation of real-time MPC in conjunction with PSO. He et al. [103] employed Hidden Markov Models (HMM) to predict driving intent, adjusting torque accordingly. Zhang et al. [104] combined adaptive cubic exponential prediction with dynamic programming to optimize torque distribution. Xu et al. [105] focused on optimizing braking force distribution in in-wheel motors for four-wheel-drive vehicles. Additionally, Wei et al. [12] introduced region-based MPC to balance battery temperature and maximize energy recovery. Basrah et al. [106] combined both linear and nonlinear MPC to enhance ABS wheel slip control and extend the regenerative braking range.

The core of MPC is the optimization of the control inputs over a prediction horizon by minimizing a cost function subject to system constraints. The classic cost function can be written as:

$$J=\sum _{k=0}^{N-1}(||y_k-y_{\mathit{ref},k}{||}_Q^2+||u_k{||}_R^2)$$

(3)

In this formula, the objective is to minimize the difference between the predicted output $y_k$ and the reference output $y_{\mathit{ref},k}$, which represents the desired system behavior. The first term in the cost function penalizes the error in tracking the reference, and the second term penalizes the control input $u_k$, ensuring that the control effort is not excessively large. The matrices Q and R are tuning parameters that control the trade-off between the two objectives: tracking performance and energy efficiency. This cost function is typically minimized over a finite prediction horizon N, and the resulting control inputs $u_k$ are implemented in real time. By minimizing this cost function, MPC can optimize the control inputs (such as braking torque) to meet both the performance and efficiency goals while ensuring system stability.

In conclusion, Model Predictive Control has proven to be a powerful tool for improving both energy recovery efficiency and system stability in regenerative braking systems, particularly in dynamic and complex conditions. By continuously predicting future states and optimizing control inputs in real time, MPC enhances the overall performance of regenerative braking systems. When combined with other advanced control and optimization techniques, MPC offers a robust and reliable solution for intelligent energy management in electric and hybrid vehicles.

5.4. Sliding Mode Control Method

SMC is a highly effective nonlinear control technique that has found widespread application in regenerative braking systems for EVs. As depicted in Figure 14, SMC operates by designing a sliding surface (in SMC, a predefined hyperplane in the state space onto which the system dynamics are forced and then maintained, defining the desired system behavior) and employing switching control to suppress nonlinear dynamics within the system, making it particularly well-suited for handling system uncertainties. This method offers robust performance, particularly in optimizing both energy recovery and system stability in EVs under various operating conditions.

In Figure 15, He et al. [107] introduced a combined fuzzy control and SMC approach for ABS. The sliding surface was constructed based on tire longitudinal slip error, and a switching function was designed, resulting in significant improvements in vehicle stability and safety. In comparison to traditional PID control, the FSMC approach effectively reduced braking distances and enhanced slip rate control accuracy. Mei [54] proposed an adaptive FSMC that dynamically adjusts wheel speed error and its rate of change, optimizing brake torque distribution. This strategy significantly enhanced both energy recovery and dynamic response performance. Liu et al. [108] presented an adaptive braking strategy that integrated SMC with vehicle mass and slope estimation, maintaining high recovery efficiency and ensuring smooth braking under varying load and slope conditions. Zhang and Cai [109] employed slope information within an SMC framework for dynamic brake force distribution between the front and rear axles, optimizing both energy recovery and system stability, particularly under city downhill cruise conditions.

Venkatesh et al. [110] introduced SMC based on Cyber-Physical Systems (CPS), improving wheel speed and braking torque distribution. This enhancement resulted in considerable improvements in system stability and recovery capacity, particularly in complex scenarios. Rajendran et al. [111] proposed an Intelligent SMC system designed to prevent overcharging of the battery through intelligent logic, leading to improved energy recovery and braking stability. Guo et al. [112] combined FSMC to develop a novel ABS performance evaluation method, which optimized the distribution of regenerative and hydraulic braking forces, thereby reducing tire slip.

Zhu et al. [113] applied SMC in a SRM drive system, integrating multi-objective optimization (MOOS) to enhance brake force distribution. This approach resulted in significant improvements in recovery efficiency, braking comfort, and battery life. Geraee et al. [114] optimized switching modes using an improved Direct Torque Control (DTC) method in conjunction with adaptive control, which improved system response speed, reduced torque fluctuations, and extended battery life.

The classic formula for designing the sliding surface s(t) in SMC is given by:

$$s(t)=Cx(t)+\mathsf{\lambda }u(t)$$

(4)

where $s(t)$ is the sliding surface, C is a matrix that defines the desired system behavior, x(t) is the state vector of the system, λ is a tuning parameter, u(t) is the control input. In this formula, the sliding surface s(t) is designed to represent the difference between the current state and the desired system behavior. The control input u(t) is adjusted in such a way that the system trajectories are forced onto this sliding surface and maintained there. The matrix C represents the desired behavior in the state space, and λ is a tuning parameter that controls the dynamics of the sliding mode. Once the system state enters the sliding surface, it is driven to remain there, ensuring the robustness and stability of the system despite uncertainties and external disturbances.

In conclusion, Sliding Mode Control offers distinct advantages in regenerative braking systems for electric vehicles. By integrating SMC with fuzzy control, intelligent logic, and multi-objective optimization, the method not only improves energy recovery efficiency but also significantly enhances the robustness and adaptability of the system. As such, SMC has emerged as a pivotal direction for future developments in EV braking control technology.

5.5. Adaptive Control Method

Adaptive control methods dynamically adjust control strategies in real time, responding to changes in driving conditions and environmental factors. These techniques have gained widespread application in the regenerative braking systems of EVs and HEVs. As depicted in Figure 16, the controller processes input parameters—such as deceleration, torque demand, speed, and SOC of the battery—optimizing the distribution of braking forces to maximize energy recovery, enhance braking efficiency, and ensure both driving comfort and safety.

Arun Kumar et al. [115] proposed a hybrid control strategy combining the ANFIS with PID control for regenerative braking in Brushless DC motors. This method significantly improved both energy recovery and driving comfort, particularly under complex environmental conditions. Kubaisi [116] utilized a multi-dimensional hidden Markov model to infer driving intent and adaptively adjust braking strength, thereby enhancing driving range and preventing the overuse of regenerative braking. Akhila and Ratnan [117] integrated ANFIS with PID control to optimize braking torque and recovery efficiency, achieving significant improvements in response speed. Han et al. [118] introduced an adaptive cooperative control strategy for front-wheel-drive HEVs, improving lateral stability and preventing excessive regeneration that could lead to slipping. This was accomplished by balancing front-to-rear wheel torque distribution and incorporating sliding mode control.

Singh et al. [119] proposed an ANFIS-based Equivalent Consumption Minimization Strategy (ECMS) to optimize power distribution between the Internal Combustion Engine and the Electric Motor, which resulted in enhanced regenerative energy recovery. Sun et al. [120] developed a dual-mode regenerative braking strategy based on Active Disturbance Rejection Control, effectively balancing charging safety and efficient energy recovery through current feedback and torque control. Arslan et al. [121] utilized Adaptive Terminal Sliding Mode Control to manage fuel cells and supercapacitors, maintaining system stability under load variations and improving overall system efficiency. Guo et al. [122] integrated Adaptive Cruise Control with the regenerative braking system, utilizing Adaptive Fuzzy Sliding Mode Control, which excelled in both car-following control and energy recovery.

In adaptive control, the control law is adjusted based on real-time estimation of system parameters. A general form of the adaptive control law for regenerative braking systems is given by:

$$u(t)=K\widehat{\theta }(t)x(t)$$

(5)

where u(t) is the control input (braking torque), (t) is the estimated parameter vector, x(t) is the state vector (system’s state), K is a gain matrix that ensures the system stability. In this formula, the control input u(t), which in the case of regenerative braking would be the braking torque, is computed by multiplying the estimated parameters $\widehat{\theta }(t)$ with the state vector x(t). The parameters $\widehat{\theta }(t)$ are adjusted in real time based on the observed performance and changes in the system, which makes the control strategy adaptive. The gain matrix K is chosen to stabilize the system, ensuring that the braking torque optimizes energy recovery while maintaining system stability and safety.

In conclusion, adaptive control methods significantly enhance regenerative braking energy recovery efficiency by dynamically adjusting control strategies in response to varying driving conditions. These methods not only extend driving range but also increase the robustness and reliability of the system. As intelligent control technologies evolve, the integration of adaptive control with fuzzy logic, sliding mode control, and other advanced techniques will continue to drive the development of more efficient and intelligent EV braking systems.

5.6. Learning-Based Control Methods

With the continuous evolution of EVs and HEVs, learning-based control methods have emerged as a pivotal technology for optimizing regenerative braking energy recovery. These methods leverage various machine learning and reinforcement learning algorithms to adaptively adjust control strategies based on changing driving environments and driver behaviors, thereby improving system performance and energy recovery efficiency.

In recent years, reinforcement learning techniques have made substantial strides in optimizing regenerative braking systems for EVs. Wu et al. [123] proposed a multi-objective optimization approach based on the Munchausen Prioritized Experience Soft Actor-Critic (MPE-SAC) algorithm. This method optimizes both braking energy recovery and battery lifespan degradation. By using Prioritized Experience Replay (PER) and emphasizing recent experience (ERE), the approach demonstrated an 8.57% improvement in energy recovery rewards, offering better adaptability and faster convergence compared to traditional methods. Yin et al. [124] introduced a Q-learning-based regenerative braking control strategy that transforms battery energy from the time domain to the space domain, optimizing energy balance during braking and enhancing energy recovery efficiency.

For Intelligent Connected Vehicles, Qiu et al. [125] developed an Intelligent Regenerative Braking Control Strategy, which accounts for the coupling effects of driving conditions and driver behavior. This strategy dynamically adjusts recovery braking force, optimizing energy recovery while improving both driving stability and vehicle economy. Li et al. [126] combined driving mode recognition with fuzzy control to propose an energy regulation management strategy for electro-hydraulic hybrid vehicles, significantly enhancing system efficiency and stability.

In addition to reinforcement learning and Q-learning, machine learning methods have also been applied effectively in regenerative braking control for EVs. Si et al. [127] developed an energy management strategy for flywheel hybrid electric vehicles, integrating Learning Vector Quantization (LVQ) neural networks and dynamic programming (DP) to optimize torque distribution. This method not only improved energy recovery efficiency but also reduced CO₂ emissions, as shown in Figure 17. He et al. [52] proposed an optimized regenerative braking strategy for dual-motor-driven EVs, combining torque optimization and electro-hydraulic coordination to significantly increase recovery efficiency, with simulation results validating the effectiveness of the strategy.

On the front of intelligent control strategies, Gupta et al. [11] introduced an intelligent regenerative braking control strategy based on friction coefficient estimation, combining FLC with an ANFIS. This strategy optimizes both energy recovery efficiency and SOC control. Prasanth et al. [128] applied machine learning techniques, utilizing multiple regression and random forest regression algorithms to optimize energy recovery during regenerative braking. Their approach outperformed traditional fuzzy logic and neural network methods, achieving a 59% improvement in energy recovery efficiency.

The use of multi-agent deep reinforcement learning (MADRL, a subset of RL where multiple autonomous agents learn to make decisions in a shared environment, often collaborating or competing) methods has also expanded within EV energy management and control strategies. As shown in Figure 18, Gan et al. [129] proposed a multi-objective cooperative control strategy based on Multi-agent Deep Deterministic Policy Gradient (MADDPG). This approach enables multiple agents to handle regenerative braking, energy distribution, and four-wheel-drive torque, optimizing hybrid electric vehicle systems and significantly improving energy management efficiency and control effectiveness. Min et al. [130] employed reinforcement learning to enhance deceleration planning algorithms, leading to a 9.62% increase in energy recovery efficiency according to their simulation results.

Recent advancements in Deep Reinforcement Learning (DRL) have also shown promise in regenerative braking systems for dual-motor EVs. Peng and He [13] developed a DRL-based regenerative braking control strategy, improving sample efficiency through an enhanced Prioritized Experience Replay mechanism. This strategy effectively addresses the exploration-exploitation dilemma, outperforming traditional methods in terms of energy recovery efficiency, braking safety, and comfort. For hydraulic hybrid EVs, Ning et al. [68] introduced a fuzzy Q-learning (FQL)-based regenerative braking optimization algorithm. This algorithm, through reinforcement learning, optimizes braking force distribution, significantly improving energy recovery efficiency.

In the domain of mining EV dump trucks, Weiwei et al. [131] proposed a DRL optimization strategy based on SAC and DDPG algorithms, significantly improving both energy efficiency and battery life. The SAC-based strategy outperformed traditional SAC control approaches, enhancing convergence speed by 166.7%. Maia et al. [132] further refined regenerative braking optimization through reinforcement learning by improving the fuzzy logic model. Chen et al. [69] applied a Fuzzy-TD3 deep reinforcement learning algorithm for electro-hydraulic composite braking coordination, optimizing the integration between regenerative and hydraulic braking systems, leading to marked improvements in braking performance and energy recovery efficiency.

Li et al. [133] proposed a game-theory-based regenerative braking control strategy for EVs. This approach optimizes the regenerative braking coefficient (K) using game theory, yielding an 18.06% improvement in energy recovery rate in NEDC simulations. Compared to traditional cruise control and fuzzy control strategies, this method ensured greater braking stability and comfort. Wu et al. [134] employed TD3 deep reinforcement learning to develop an energy management strategy (EMS) for plug-in hybrid electric vehicles (PHEVs), optimizing driving mode selection and torque distribution. Experimental results indicated that this strategy outperformed conventional DRL-based EMS approaches in terms of energy efficiency.

Min et al. [135] proposed an online parameter learning-based vehicle deceleration prediction model for optimizing autonomous regenerative braking control in EVs. By adjusting the regenerative braking strategy in real time based on individual driving characteristics, this model significantly enhanced energy recovery efficiency and driving comfort. Xu et al. [136] employed a DRL-optimized regenerative braking strategy, utilizing the DDPG (Deterministic Policy Gradient) algorithm for energy management, which notably improved convergence speed and demonstrated real-world application advantages.

In reinforcement learning-based control, the objective is to maximize the cumulative reward over time. The classic objective function for the RL problem can be expressed as:

$$J=\sum _{t=0}^T{\gamma }^t{r}_t$$

(6)

where J is the total cumulative reward, γ is the discount factor (which determines the importance of future rewards), $r_t$ is the reward at time step t, T is the total number of time steps. In this formula, the objective is to maximize the total cumulative reward J over time. The reward at each time step $r_t$ depends on the actions taken by the agent (in this case, the control strategy for regenerative braking). The discount factor γ determines how much future rewards are valued compared to immediate rewards. A higher γ means that future rewards are more heavily weighted. This objective is typically optimized using algorithms such as Q-learning, Deep Q-Networks (DQN), or more advanced techniques like TD3 and SAC, depending on the problem at hand.

In conclusion, learning-based control methods, particularly reinforcement learning, have made significant progress in optimizing regenerative braking for EVs and HEVs. These methods, by adapting control strategies in real time based on driving conditions and driver behavior, enhance energy recovery efficiency, system robustness, and overall vehicle performance. As intelligent control algorithms continue to advance, their integration will play a crucial role in further optimizing regenerative braking technologies.

6. Comparison of Intelligent Methods and Their Application in 4WID EVs

6.1. Overview and Comparison of Intelligent RB Control Strategies

Six intelligent control strategies—FLC, ANN, MPC, SMC, Adaptive Control, and Learning-based Control—each possess distinct characteristics and offer various advantages and limitations in the context of regenerative braking energy recovery. The following provides a comparative overview of these methods:

• Fuzzy Logic Control.

FLC is grounded in experience-based rules and does not rely on precise mathematical models, making it particularly adept at managing nonlinear, strongly coupled, and time-varying systems. Under typical urban driving conditions, FLC can achieve energy recovery rates ranging from 15% to 30%, with certain studies indicating up to a 5% improvement in recovery efficiency over traditional fixed-ratio control strategies [68,137]. Key advantages of FLC include its simple structure, strong real-time capabilities, and low implementation cost. However, its control precision is constrained by manually designed rule bases and membership functions, and it lacks self-learning and adaptability [15].

2.

Artificial Neural Network Control.

ANNs offer powerful nonlinear modeling and self-learning capabilities, enabling the system to autonomously approximate optimal braking strategies after sufficient training. With adequate data, ANN can achieve energy recovery efficiencies ranging from 8% to 40%; for instance, a Fuzzy Neural Network-based self-learning control system achieved 39.61% energy recovery under FTP-75 conditions [138,139]. ANN control is particularly suited for dynamic control in complex conditions due to its ability to respond quickly, output continuously, and maintain smooth operation. However, its generalization ability is highly dependent on the distribution of the training data, and its robustness outside the training domain needs further enhancement. Additionally, the initial training cost is relatively high [140].

3.

Model Predictive Control.

MPC operates within a rolling optimization framework, considering prediction models and constraint conditions, which facilitates multi-objective coordinated control. Research has demonstrated that MPC can improve regenerative energy recovery efficiency by 5% to 8% compared to traditional strategies, and reduce vehicle energy consumption by as much as 17% [8,141]. MPC excels at optimizing energy efficiency, braking stability, and comfort within a unified control framework, making it particularly suitable for high-end electric vehicle platforms. However, MPC’s main drawback is its significant computational resource requirement for solving online optimization problems during each control cycle, often necessitating model simplifications or fast solvers [75].

4.

Sliding Mode Control.

SMC is widely employed in challenging scenarios such as low-adhesion braking and Anti-lock Braking System (ABS) control due to its robustness and capacity to handle uncertainties. By utilizing the invariance principle, SMC stabilizes the slip ratio, thereby achieving dual objectives of braking stability and energy recovery. Studies indicate that SMC effectively maintains the optimal slip ratio under extreme conditions, preventing wheel lock-up [142]. However, SMC may induce high-frequency oscillations, which can affect driving comfort. To mitigate this, fuzzy control is often integrated to adjust the approach law and enhance comfort [143].

5.

Adaptive Control.

Adaptive control adjusts the controller structure or gain in real time through online system parameter identification, enabling quick responses to environmental and vehicle state changes, such as variations in SOC, motor power degradation, or road surface friction. It does not require extensive offline training data, making it well-suited for platforms with highly variable conditions. Literature suggests that adaptive control can effectively improve braking smoothness and system stability, while maintaining high energy recovery rates across different SOC conditions [144]. However, adaptive control lacks predictive capabilities for future states, and its performance is primarily reliant on the design of the adjustment strategy.

6.

Learning-based Control.

Learning-based control methods, including reinforcement learning, genetic algorithms, and deep neural strategies, learn near-optimal control policies through extensive interactions or simulation data. For example, a four-motor distribution strategy optimized with a genetic algorithm can achieve a 22.8% increase in recovery efficiency and a 4.8% improvement in braking stability, while reinforcement learning strategies can achieve near-global optimal performance in specific scenarios, such as reaching 99% of dynamic programming results [6,145]. Learning-based methods are highly effective for high-dimensional, strongly coupled systems and offer long-term self-improvement potential. However, they are highly data-dependent and resource-intensive during the training phase, with complex development and debugging processes. Once trained, these control strategies require minimal computational resources, making them well-suited for real-time deployment [15].

To better compare the performance of various intelligent control algorithms in regenerative braking systems (RBS), we provide specific quantitative targets for energy recovery efficiency, braking stability, and computational cost for each control strategy:

• Energy Recovery Efficiency:

Fuzzy Logic Control (FLC): Achieves energy recovery rates ranging from 15% to 30%, with up to 5% improvement over traditional fixed-ratio control strategies under urban driving conditions. Artificial Neural Network (ANN): Recovery efficiency can vary from 8% to 40%, with 39.61% recovery achieved under FTP-75 conditions, offering significant improvements in dynamic control for complex conditions. Model Predictive Control (MPC): Improves energy recovery efficiency by 5% to 8% compared to traditional methods, and can reduce vehicle energy consumption by up to 17% in optimized scenarios. Sliding Mode Control (SMC): Enhances recovery efficiency by 2.8% and ensures stability by maintaining optimal slip ratios under extreme conditions such as low-adhesion surfaces. Adaptive Control: Provides up to 6.2% improvement in energy recovery efficiency across varying driving conditions, particularly in fluctuating load scenarios. Learning-based Control (including DRL and Q-learning): Offers more than 10% improvement, with some methods reaching 59% improvement in energy recovery efficiency under urban driving conditions.

2.

Braking Stability:

FLC and ANN: Provide moderate stability, particularly in urban and mixed driving conditions, though their precision may decrease during extreme braking events. MPC and SMC: Excel in braking stability, particularly in emergency braking situations, and ensure consistent system performance under dynamic and high-stress conditions. Adaptive Control: Highly effective in maintaining stability, with real-time adjustments ensuring robust performance even under variable road and environmental conditions. Learning-based Control: Demonstrates optimal braking stability when integrated with other algorithms, ensuring seamless transitions between regenerative and mechanical braking.

3.

Computational Cost:

FLC and SMC: These methods have low computational cost and are suitable for real-time deployment on platforms with limited hardware, making them ideal for production vehicles with fewer computational resources. ANN: Requires significant training time and computational resources during the learning phase. However, once trained, the real-time computational cost is moderate. MPC: Demands high computational resources due to real-time optimization and model predictive capabilities, making it more suitable for advanced EV platforms with powerful hardware. Learning-based Control: Data-intensive during training, but once trained, the real-time cost can be minimized. However, the complexity of these methods may still require higher computational resources compared to simpler control strategies.

As shown in

Table 5. Comparison of Intelligent Control Algorithms on Key Performance Indicators.

Algorithm Category	Recovery Efficiency	Stability	Smoothness	Real-Time	Implementation Complexity	Computational Cost	Scalability
FLC	4	4	4	5	5	5	4
ANN	5	4	4	5	3	3	3
MPC	3	5	5	2	2	4	3
SMC	4	5	3	5	4	4	4
Adaptive	4	4	4	5	3	4	4
Learning	5	5	4	5	1	2	3

comparing algorithms across five key performance indicators, the performance scores range from 1 to 5, with 5 indicating the best performance. These quantitative metrics are defined as follows: Recovery Efficiency indicates the effectiveness of regenerative braking energy recapture; Stability reflects the ability to maintain vehicle safety and suppress disturbances; Smoothness assesses driving comfort and transition quality; Real-Time performance evaluates responsiveness and hardware deployment feasibility; Implementation Complexity rates the ease of development and tuning; Computational Cost refers to runtime resource demands; and Scalability indicates adaptability to different system configurations and operating conditions.

As shown in Figure 19, in terms of energy recovery efficiency, FLC and ANN methods perform well in the mid-to-high range (15–40%), while learning-based control strategies have the potential to approach near-global optimization. MPC offers superior overall energy efficiency through multi-objective constraint optimization, making it particularly effective in dual-motor four-wheel drive (4WID) systems. For braking stability and safety, SMC stands out due to its excellent disturbance suppression and slip rate control, making it particularly suitable for ABS integration, while adaptive control excels in adjusting regenerative/friction braking ratios based on varying braking conditions. FLC and ANN provide moderate robustness, while MPC, SMC, and learning-based control strategies achieve optimal performance under specific constraint conditions. In terms of braking smoothness and driving comfort, MPC excels at maintaining smoothness, avoiding abrupt pedal changes, while FLC ensures seamless switching and ANN offers continuous output. SMC, however, may impact comfort due to oscillatory behavior, requiring adjustments to the approach law to alleviate this issue.

Regarding algorithm implementation complexity, FLC and SMC are relatively simple and can be easily embedded into real-time control platforms. ANN and learning-based methods require complex training processes, but once trained, they incur lower computational burdens during operation. MPC demands significant computational resources due to its online optimization process, making it more suitable for platforms with adequate hardware capabilities. In practice, achieving an optimal balance between performance gains and implementation costs is essential [146,147].

6.2. Application of Intelligent Algorithms in RB Energy Recovery of 4WID EVs

With the rapid advancement of EV technology, four-wheel independent-drive (4WID) EVs have demonstrated distinct advantages in energy recovery due to their flexible drive and braking capabilities. However, optimizing braking energy recovery efficiency while ensuring vehicle safety and driving comfort has become a central focus of current research. In recent years, intelligent control strategies have made substantial progress, with algorithms such as FLC, ANN Control, MPC, SMC, Adaptive Control, and Learning-based Control being widely applied to the braking energy recovery systems of 4WID EVs, yielding significant improvements.

FLC, known for its robust fault tolerance and ability to adapt to system nonlinearities, is frequently employed in braking energy recovery distribution. Yin et al. [137] proposed an energy recovery strategy based on fuzzy control, which resulted in a 9.8% improvement in energy recovery under urban driving conditions, while also reducing tire slip and enhancing driving smoothness. Similarly, Ning et al. [68] incorporated fuzzy Q-learning into a hydraulic hybrid braking system, which led to a 12% reduction in braking energy loss.

ANN control leverages its self-learning and nonlinear modeling capabilities to efficiently predict torque distribution under complex driving conditions. Li et al. [138] applied a PSO neural network controller, achieving a 2.4% improvement in energy recovery and significantly enhancing vehicle lateral stability. Jamil et al. [139] integrated ANN with adaptive fuzzy control, and experimental results demonstrated a 6.2% improvement in efficiency under mixed urban and highway driving conditions.

MPC optimizes braking in real time by forecasting future driving conditions and vehicle states. Zheng et al. [141] developed an adaptive MPC algorithm that reduced energy loss by 15% under extreme braking conditions compared to traditional PID controllers, while maintaining vehicle safety and stability. In a similar vein, Xu et al. [8] combined MPC with SMC to further enhance energy recovery and vehicle safety, particularly when road surface adhesion coefficients varied.

SMC is renowned for its robustness and capacity to resist disturbances, making it particularly effective in managing parameter variations and system uncertainties. Sun et al. [142] applied an SMC strategy that increased recovery efficiency by 2.8%, significantly improving vehicle yaw stability. Zhao et al. [143] integrated neural network-based sliding mode control to coordinate ABS and energy recovery, which not only reduced braking distances but also improved energy recovery efficiency.

Adaptive control methods dynamically adjust control parameters to optimize torque distribution based on real-time conditions. Liu et al. [144] designed a model-free adaptive controller capable of accurately performing braking force distribution tasks, even under extreme conditions such as fluctuations in battery voltage. This adaptability enhances the system’s overall performance under variable operational circumstances.

Learning-based control methods, such as reinforcement learning, optimize energy recovery strategies through long-term experience accumulation. Debella et al. [145] demonstrated that a torque distribution strategy driven by reinforcement learning could increase energy recovery by more than 10%. Vodovozov et al. [15] emphasized the scalability and practical potential of this method in multi-motor, multi-wheel drive systems, highlighting its ability to improve the overall energy recovery process.

In conclusion, intelligent algorithms significantly enhance the energy recovery efficiency of regenerative braking systems in 4WID electric vehicles, while also improving system safety and stability. Each intelligent control method offers unique strengths in various practical applications, with some excelling in energy recovery, while others focus on enhancing vehicle stability and driving comfort.

Table 6. Comparison of the Performance of Intelligent Algorithms in RBS in 4WID EVs.

Algorithm Type	Main Advantages	Performance Improvement	References
Model Predictive Control	Handles non-linearity and uncertainty	Energy recovery improved by 9.8%	[137]
Neural Network Control	Prediction and self-learning capability	Energy recovery improved by 2.4%	[138]
Model Predictive Control	Global optimization, adaptive to conditions	Energy loss reduction by 15%	[141]
Sliding Mode Control	Robust, disturbance-resistant	Energy recovery increased by 2.8%	[142]
Adaptive Control	Real-time adjustment of control parameters	Efficiency improved by 6.2%	[139]
Learning-based Control	Continuous self-optimization	Energy recovery increased by 10%+	[145]

summarizes the performance improvements achieved by typical intelligent algorithms, providing valuable insights for the optimization of future control strategies and system integration.

7. Challenges and Development Trends of RBS in EVs

7.1. Current Key Challenges

The RBS in EVs still faces a range of challenges in practical applications. As shown in Figure 20, first, energy recovery efficiency is limited. At low speeds or during emergency braking, the regenerative torque from the motor is insufficient, requiring mechanical friction braking intervention. Road conditions, vehicle speed, and load also significantly impact the effect. Currently, vehicles can typically recover only 10% to 30% of the energy consumed during driving, which is far below the theoretical value. Overcoming ineffective recovery at low speeds and the energy waste caused by variable driving conditions is a pressing issue [52,148].

Secondly, battery aging and thermal management pose limitations. Frequent regenerative braking leads to large current charging, which can cause battery heating and lifespan degradation, particularly at high SOC or low temperatures. The battery’s ability to accept energy is limited, and the battery management system (BMS) often restricts the feedback power to protect the battery. The accurate estimation of battery state parameters is also affected by aging and environmental factors. Therefore, thermal management must be coordinated with the recovery strategy to enhance efficiency while mitigating the high current impact on battery life and safety [41]. Moreover, the control strategy must evolve from merely avoiding immediate dangers (like overcharging) to actively minimizing cumulative degradation per unit of energy recovered. This necessitates a deeper integration between the BMS and the braking controller, considering the battery’s state of health (SOH) in real-time optimization.

Thermal management under high-power regeneration is another pivotal challenge. The substantial heat generated in the motor (acting as a generator) and the battery during aggressive braking can lead to rapid thermal saturation. To prevent damage, the system is forced to derate the regenerative torque or shut it down entirely, creating a significant gap between peak and sustainable recovery power. Effective heat dissipation is therefore paramount. Advanced thermal management systems (TMS)—featuring direct oil cooling for motors and refrigerant-cooled battery packs—are essential to maintain components within their optimal temperature range, thereby extending the duration of high-power regeneration. Furthermore, predictive thermal management, leveraging preview information from navigation or V2X systems, could pre-cool critical components ahead of anticipated braking events, unlocking a higher and more consistent portion of the system’s recovery capability [149].

Next, brake coordination and stability issues are prominent. RBS needs to distribute braking force in real time between the electric motor and hydraulic brake systems, satisfying the driver’s deceleration needs, maximizing recovery, and ensuring vehicle stability. In emergency braking or low-traction road conditions, improper control can lead to wheel lock-up or skidding. Achieving smooth and safe seamless switching remains a challenge [92].

As energy, safety, and comfort constraints coexist, the complexity of control strategies and model precision issues emerge. MPC and intelligent algorithms are introduced, but the strong nonlinearity and time-varying parameters of the vehicle and battery systems make it difficult to obtain high-fidelity models. Model deviations weaken the control effect, simplified models reduce effectiveness, and high-precision models increase computational load. Sensor noise further limits robustness. In terms of intelligent control algorithms, fuzzy control, neural networks, reinforcement learning, etc., show potential in simulations, but practical applications face challenges such as complex parameter tuning, lack of theoretical support, high computational requirements, and poor generalization ability, making it difficult to meet functional safety and real-time requirements [91].

Finally, 4WID EVs bring high-dimensional control challenges. Multi-motor architectures increase recovery potential but require high-dimensional optimization of braking forces at each wheel. This must not only maximize recovery but also ensure front-rear axle balance and yaw stability. Real-time solution and torque coordination significantly increase algorithm complexity and computational load. Multi-motor failures and complex dynamics further add uncertainty. Although composite control strategies have been proposed, implementing efficient and robust high-dimensional coordinated control on a whole vehicle remains a difficult task [9,10].

7.2. Future Development Trends

Looking ahead, RBSs in EVs are advancing toward more sophisticated and integrated vehicle coordination systems. As illustrated in Figure 21, the key future trends include the following:

• Brake-by-Wire Technology.

Brake-by-wire technology replaces traditional hydraulic brake lines with electrical signals, enabling independent and precise braking force distribution to each wheel. This system lays the foundation for seamless integration of regenerative and friction braking. Experimental studies suggest that brake-by-wire systems can improve energy recovery efficiency by 5–10% compared to conventional electro-hydraulic systems, while also reducing braking distance by up to 8% under low-adhesion conditions [17]. Smart BMS technology will dynamically adjust feedback current based on SOC, temperature, and power history. Additionally, incorporating State of Health (SOH) models will optimize the trade-off between energy recovery and battery life, reflecting a holistic approach to braking, driving, and energy storage design [100].

• Control Algorithms Evolving from Empirical Rules to Artificial Intelligence.

DRL will allow systems to learn optimal composite braking strategies autonomously. For instance, DRL-based strategies have been shown to achieve energy recovery rates of 35–40% in urban driving cycles, approaching 99% of the theoretical global optimum [13,145]. Hybrid algorithms that combine fuzzy-neural networks and adaptive MPC will strike a balance between robustness and optimization performance, continuously adapting to challenges like battery aging, road adhesion variations, and driver behavior.

• Integration of Vehicle-to-Everything (V2X) and Autonomous Driving.

The introduction of external information flows, such as V2X, will enable predictive energy management strategies. Vehicles will utilize V2X, high-definition maps, and GPS to anticipate deceleration conditions, planning long-distance coasting in advance. This could increase regenerative energy recovery by 15% to 40% in urban environments, with Vehicle-to-Vehicle (V2V) and Vehicle-to-Road (V2R) communication potentially boosting this to 50% [150]. Moreover, such systems can improve braking stability by reducing yaw rate deviation by up to 20% in emergency scenarios [145].

• Multi-Objective Optimization Frameworks.

Future developments will incorporate multi-objective optimization frameworks to harmonize energy efficiency, battery degradation, braking safety, and smoothness within a single model. Advanced hierarchical MPC or multi-objective DRL frameworks have demonstrated the ability to improve overall energy recovery by 10–15% while maintaining braking comfort (e.g., jerk < 10 m/s³) and reducing battery degradation by 5–8% over the vehicle lifespan [10,145]. These systems will collaborate with ESP, Autonomous Emergency Braking (AEB), and other longitudinal and lateral control systems to enable customized, scenario-based strategies.

• Hardware Advancements.

The integration of HESS, such as supercapacitors, flywheels, and lithium batteries, coupled with bidirectional DC-DC converters, will help absorb peak power, smooth current flow, and reduce heating. Studies show that HESS can improve regenerative energy recovery by 8–12% and extend battery cycle life by 15–20% compared to battery-only systems [30,61]. As the cost of power electronics and high-energy-density components decreases, HESS and integrated motor-brake units will become more widespread, providing higher power density and reliability for RBSs.

• Standardized Testing and Technology Maturity.

The establishment of a unified evaluation system for regenerative braking systems is critical. This system would assess energy recovery efficiency, braking safety, thermal management, durability, and energy contribution in autonomous driving scenarios. Proposed metrics include energy recovery ratio (targeting > 35% in urban cycles), braking stability index (e.g., yaw rate error < 0.5°/s), and thermal safety margin (battery temperature < 45 °C during aggressive braking) [146]. Additionally, supporting hardware-in-the-loop and road-testing platforms will facilitate cross-scenario quantification and the development of regulatory standards [151].

In summary, regenerative braking systems are evolving through a closed-loop trajectory that integrates system optimization, intelligent control, scenario prediction, hardware advancements, and standard evaluations. These advancements will enhance energy efficiency, vehicle safety, and battery longevity, offering electric vehicles improved range, lower costs, and increased reliability.

8. Conclusions and Outlook

This review systematically summarizes and analyzes the key technologies and research progress of RBS in EVs. By exploring various aspects such as system architecture, traction motors, energy storage units, drive configurations, brake actuators, intelligent control algorithms, and evaluation systems, this paper highlights the latest technological trends in the RBS field and presents the following key conclusions:

Key Conclusions:

• Highly Integrated System Architecture is Fundamental to Enhancing Efficiency.

The evolution toward a “predict–control–execute” closed-loop architecture integrates multi-source sensor data to enable real-time optimization of braking force distribution between regenerative and hydraulic systems. This integrated approach—combining regenerative-priority braking, decoupled pedal systems, and brake-by-wire technologies—not only improves energy recovery but also ensures braking safety and consistent pedal feel. This addresses the first contribution outlined in the introduction by providing a holistic system-integration perspective that clarifies the interaction mechanisms across RBS components, thereby offering a cohesive framework that supports both current performance demands and future expansion into autonomous driving.

2.

High-Power-Density Motors and Hybrid Energy Storage Systems Define the Recovery Potential.

PMSMs are predominant in modern EVs due to their high efficiency and power density, while IMs and SRMs serve niche applications. However, the energy recovery ceiling is constrained by the limitations of lithium-ion batteries under high-power regenerative charging. HESS, combining batteries with supercapacitors or flywheels, alleviate these constraints by managing power peaks and improving battery longevity. This aligns with the system-integration theme by highlighting how component-level advances and hybrid configurations directly impact the upper limits of recoverable energy.

3.

4WID Architectures Significantly Enhance Energy Recovery and Dynamic Control.

The 4WID configuration maximizes regenerative braking potential by enabling individual wheel control and all-wheel energy recovery. Compared to conventional single- or dual-axle layouts, 4WID systems demonstrate superior energy recovery efficiency—up to 30–38%—and enhanced vehicle stability under diverse driving conditions. This quantitatively substantiates the third contribution of the paper, clearly demonstrating the advantages of 4WID in both energy recovery and dynamic performance, thereby addressing a noted gap in prior reviews that lacked explicit 4WID-focused analysis.

4.

Intelligent Control Algorithms Are Diversifying Toward Hybrid and Learning-Based Methods.

A comparative analysis of six categories of intelligent algorithms—Fuzzy Logic, Neural Networks, Model Predictive Control, Sliding Mode Control, Adaptive Control, and Learning-Based methods—reveals that each has distinct strengths in terms of efficiency, stability, real-time capability, and implementation complexity. Hybrid strategies that combine model-based and data-driven approaches are increasingly effective in managing high-dimensional control problems in 4WID systems. This directly responds to the second contribution by providing a systematic comparison of intelligent control methods, summarizing their suitability for varying scenarios, and underscoring the trend toward hybrid and learning-enabled architectures.

5.

Multi-Objective Optimization and Standardization Remain Critical Challenges.

Future RBS developments must concurrently address energy efficiency, braking safety, drivability, thermal management, and battery lifespan. The absence of unified testing standards and evaluation frameworks—especially for emerging scenarios such as vehicle-to-everything (V2X) communication and autonomous driving—hampers direct comparison and industrial scalability. This conclusion reinforces the need for integrated evaluation frameworks as anticipated in the introduction, highlighting that overcoming these barriers is essential for next-generation RBS deployment.

Research Outlook:

• System-Level Collaborative Design Across the Entire Lifecycle.

Future RBS technologies will further emphasize the “drive-brake-storage” system-level collaborative design. By integrating motor and brake units, energy flow paths can be shortened, reducing conversion losses. The BMS will be tightly coupled with the brake controller to adjust energy recovery power adaptively, based on both battery SOH and thermal safety conditions. This integration will facilitate the co-optimization of energy efficiency and battery longevity, and thermal performance.

2.

Safety and Explainability in Intelligent Algorithm Deployment.

The application of DRL, meta-learning, and transfer learning will require a transition from simulation-based verification to hardware-in-the-loop (HIL) and real-vehicle closed-loop testing. Explainable control strategies, robust margin analysis, and compliance with ISO 26262 [152] for functional safety will be critical for large-scale deployment. Additionally, distributed computing for 4WID systems and vehicle-cloud collaborative reasoning will help alleviate onboard computational demands while maintaining real-time performance.

3.

Vehicle-to-Road Collaboration and Context Prediction.

Through the use of V2X communication, high-definition maps, and fleet information, RBS can predict road conditions (e.g., traffic lights, slopes, congestion) well in advance. By planning optimal energy trajectories for long-distance coasting and deceleration, RBS can improve energy recovery rates in urban environments while reducing tire and brake friction wear.

4.

Hybrid Energy Storage and High-Reliability Integration of Power Electronics.

As the cost of low-resistance silicon carbide (SiC) devices, bidirectional DC-DC converters, and high-power-density supercapacitors continues to decrease, HESS will become more widely applicable in mid- to high-end vehicles. The coupling of flywheels, batteries, and capacitors through multi-port inverters and integrated thermal management will improve system power density and reduce volume, providing greater installation flexibility for passenger vehicle platforms.

5.

Unified Evaluation Systems and Standards.

To further accelerate the development of RBS technologies, a multi-dimensional evaluation system should be established. This system would assess factors such as energy recovery efficiency, braking distance, stability, thermal safety, lifespan degradation, and contributions to autonomous driving scenarios. Moreover, the development of rapid testing protocols combining road testing, test benches, and HIL testing will enable the quantification of the overall benefits of new solutions, providing technical support for policy subsidies, carbon credits, and international standardization efforts.

6.

Vehicle-to-Grid (V2G) Integration and Broader Ecosystem Interaction.

Looking beyond the vehicle’s immediate energy recovery, the role of RBS should be considered within the broader energy ecosystem. With the advancement of bidirectional charging technology, the energy harvested through regenerative braking can transcend its purpose of merely extending driving range. It can be utilized as a distributed energy resource, where an EV could potentially supply stored braking energy back to the grid (V2G) during peak demand periods or to stabilize local microgrids. This evolution transforms the RBS from an onboard efficiency technology into a grid-supportive asset, adding a significant layer of practical relevance and economic potential. Realizing this vision requires future RBS controllers to be co-designed with smart charging algorithms and compliant with grid communication protocols (e.g., ISO 15118 [153]). Key research challenges include developing standards for the aggregation and valuation of regenerative energy, formulating strategies to optimize battery cycling between mobility and grid services, and comprehensively evaluating the holistic economic and environmental benefits of this integration [154].