Comparative Modeling and Experimental Validation of Two Four-Wheel Omnidirectional Locomotion Architectures for a Modular Mobile Robot

Abstract

This paper presents a comparative modeling and experimental validation study for a modular four-wheel omnidirectional mobile robot, focusing on two locomotion architectures implemented on the same platform: four omni wheels (90° rollers) and four Mecanum wheels (45° rollers). Both configurations were evaluated under identical benchmark conditions on a 1 m × 1 m square path (4 m total path length), using the same nominal 12 V supply and the same test duration, in order to ensure a fair and reproducible cross-architecture comparison. A MATLAB/Simulink–Simscape dynamic model was developed for both architectures, while experimental validation was performed using Hall-effect current sensors integrated into the drive modules. Based on the measured and simulated motor currents, a 12 V-based electrical input-power estimate was evaluated at both motor and robot level. For the considered benchmark, the four-Mecanum configuration exhibited a lower measured input-power estimate than the four-omni configuration (17.88 W vs. 25.75 W), corresponding to an approximate reduction of 30.6% under the adopted assumptions. At robot level, the deviation between simulated and measured total input-power estimate was 3.70% for the four-omni architecture and 21.42% for the four-Mecanum architecture, indicating higher predictive agreement for the omni-wheel model in its present form. The comparative analysis also suggests that wheel–ground interaction and roller geometry influence not only the measured current demand but also the level of agreement between simulation and experiment. Although the present study is limited to a single standardized benchmark and nominal-voltage conditions, it provides a controlled basis for comparing the two locomotion solutions and for identifying directions for further model refinement. The findings should therefore be interpreted as benchmark-specific comparative results offering practical guidance for locomotion architecture selection and for future refinement of friction-aware omnidirectional robot models.

1. Introduction

Omnidirectional mobile robots have attracted increasing attention in industrial, service, and assistive applications because they can generate longitudinal, lateral, and rotational motion without requiring platform reorientation. This capability is particularly advantageous in confined spaces and in tasks that demand high maneuverability and accurate trajectory tracking [1,2,3,4,5]. Among the most widely used solutions, platforms equipped with omni wheels or Mecanum wheels provide holonomic motion and operational flexibility, making them well suited for modular robotic systems intended for complex indoor environments [2,3,4,5].

Beyond their well-known kinematic advantages, omnidirectional mobile robots also raise important energy-related challenges. In practice, their energetic behavior is strongly influenced by passive roller losses, wheel–ground interaction, friction anisotropy, slip effects, payload variation, and the control effort required to execute a given motion task [6,7,8,9,10,11,12,13]. As a result, energy efficiency has become an important design criterion, since it directly affects battery autonomy, actuator sizing, thermal loading, and the practical selection of the locomotion architecture [6,9,11,12].

Recent research has addressed energy-related issues from several complementary perspectives. Practical energy consumption models have been proposed for omnidirectional mobile robots to evaluate power demand under different operating conditions, while parameter estimation approaches have been used to improve the prediction of robot power consumption in the presence of uncertain payloads or model parameters [6,9]. At the same time, current monitoring approaches based on Hall-effect sensing provide experimentally accessible data for validating the current demand and the corresponding electrical input-power estimate of drive modules in mobile robotic platforms [10]. Broader review studies have also emphasized that accurate energy prediction is becoming increasingly important for heterogeneous mobile robots, particularly when robot architecture, environment, and mission profile all influence the energetic outcome [11,12].

Another important research direction concerns the influence of wheel construction and wheel–ground interaction on motion performance and energy losses. Experimental studies on omnidirectional wheels have shown that rolling resistance under quasi-static conditions is not negligible and should be considered when evaluating efficiency [7]. For Mecanum wheels, friction-aware dynamic formulations and orthotropic contact models have demonstrated that anisotropic friction and roller geometry significantly affect force transmission and motion response [8,14]. Additional kinematic and kinetic studies on omnidirectional wheel arrangements further confirm that roller geometry influences the global behavior of the platform [15]. These findings are particularly relevant when comparing omni-wheel and Mecanum-wheel architectures because the two solutions rely on different roller orientations and therefore exhibit different mechanisms of power dissipation during identical maneuvers [7,8,15].

Another active line of research focuses on trajectory generation, motion control, and energy-aware operation. Previous studies have investigated energy-optimal trajectory design for Mecanum-wheeled robots, as well as trajectory simulation and energy consumption evaluation using MATLAB/Simulink–Simscape Multibody R2024a (MathWorks, Natick, MA, USA)-based approaches [16,17]. In parallel, several advanced control strategies have been proposed for omnidirectional robots, including adaptive integral terminal sliding mode control, kinematic control methods, optimal control, self-organizing fuzzy neural network approaches, nonlinear adaptive optimal control, online controller tuning, and robust control under time-delay conditions [18,19,20,21,22,23,24,25,26,27]. These studies clearly indicate that both the selected control strategy and the executed trajectory can influence actuator torque demand, current peaks, and overall power consumption. Additional work on the dynamic modeling and control of Mecanum-wheeled vehicles under disturbances and varying loads further highlights the importance of consistent multibody modeling when evaluating energetic performance [23]. At a broader system level, recent path-planning studies have also shown that the kinematic structure of the mobile robot can influence the solution obtained when path length and energy efficiency are considered simultaneously [28].

Although the literature has advanced significantly in modeling, control, and energy-aware motion planning, the direct power comparison between omnidirectional architectures remains insufficiently explored. Most available studies focus on a single wheel technology, a single control strategy, a single trajectory-planning method, or a single layer of the problem, such as kinematics, dynamics, or wheel–ground interaction [6,7,8,9,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32]. Consequently, systematic comparative studies between four-wheel omni-drive platforms with 90° roller orientation and four-wheel Mecanum-drive platforms with 45° roller orientation, performed under the same modeling framework and identical experimental conditions, are still limited [6,10,12,14,17,23,28,29,33].

To address this gap, this paper proposes a unified dynamic modeling and experimental validation framework for the comparative analysis of two four-wheel omnidirectional locomotion architectures implemented on the same modular mobile robot platform. A CAD-consistent MATLAB/Simulink–Simscape Multibody R2024a (MathWorks, Natick, MA, USA) model is developed for both configurations, and the comparison is performed under identical benchmark conditions, including the same nominal 12 V supply, the same test duration, and the same 1 m × 1 m square reference trajectory. The simulated current demand and the corresponding 12 V-based input-power estimate are further assessed experimentally using Hall-effect current sensing integrated into the drive modules. The contribution of the present work lies primarily in a controlled comparative modeling and experimental validation framework applied to two interchangeable four-wheel omnidirectional locomotion architectures under identical benchmark conditions rather than in the proposal of a new contact law formulation. Within this framework, the study aims to provide a fair and reproducible comparison of the two architectures and to clarify the influence of roller geometry and wheel–ground interaction on the resulting current demand, input-power estimate, and model fidelity.

2. Related Work

In recent years, research on omnidirectional mobile robots has converged around four complementary themes: the development of energy consumption models suitable for early-stage design, realistic descriptions of wheel–ground interaction and associated loss mechanisms, the design of control laws and motion strategies that directly influence power demand, and the use of dynamic simulation frameworks such as MATLAB/Simulink–Simscape Multibody R2024a (MathWorks, Natick, MA, USA) for reproducible evaluation. However, these directions are often treated separately and typically for a single locomotion architecture, which makes objective cross-architecture comparisons difficult. As a result, there remains a clear need for a controlled comparative study conducted under identical conditions to quantify the power and energy demand differences between four-omni (90° rollers) and four-Mecanum (45° rollers) architectures on a standardized benchmark such as the 1 m square trajectory considered in this work.

2.1. Energy Modeling and Power Consumption Estimation

A significant body of recent work has focused on practical methods for estimating the energy consumed by omnidirectional robots, enabling designers to anticipate power requirements and autonomy limits. Practical robot-level energy models have been proposed to define key indicators such as instantaneous power and total energy, which are essential for fair comparisons across experiments and simulations [6]. In addition, when payload conditions are uncertain or variable, parameter estimation strategies can be used to calibrate power consumption models and improve the credibility of simulation-based evaluations [9]. Although these contributions provide a solid basis for defining energy metrics and parameter identification, they do not offer a direct comparison among multiple omnidirectional locomotion architectures under fully identical assumptions regarding geometry, inertia, wheel–ground interaction, and motion profile, which is the primary objective of the present work.

2.2. Wheel–Ground Interaction and Loss Mechanisms in Omni and Mecanum Drives

Energy differences between omnidirectional architectures are strongly influenced by wheel–ground interaction losses, including slip, internal friction within rollers, and direction-dependent contact behavior. For Mecanum wheels in particular, the angled rollers create anisotropic contact conditions that significantly affect traction, torque demand, and efficiency. In this context, orthotropic-friction formulations supported by experimental validation have highlighted the strong sensitivity of actuation requirements to contact parameters [8]. Complementarily, experimental studies on omni wheels with passive rollers have quantified rolling-resistance effects that can be used to parameterize wheel losses and to interpret energy differences observed at platform level [7]. While such studies improve the realism of contact and loss modeling, they usually focus on a single wheel technology or on component-level characterization rather than on integrating multiple locomotion architectures into a unified benchmark where energy can be compared directly.

2.3. Control, Trajectory Tracking, and Energy-Aware Motion Strategies

Beyond contact modeling, motion strategy and trajectory tracking have a decisive impact on peak power, current demand, and total energy, especially for short maneuvers involving frequent direction changes. Energy-aware trajectory design has been investigated through polynomial formulations optimized with respect to energy-related criteria for Mecanum-wheeled robots [16]. From a control perspective, robust and adaptive controllers have been widely studied to improve tracking accuracy and mitigate disturbances, thereby indirectly influencing energy demand. For example, adaptive integral terminal sliding-mode control has been proposed for trajectory tracking in Mecanum-wheeled omnidirectional robots [18], while optimal control approaches based on sliding-mode concepts have also been investigated for the same class of platforms [20]. In addition, kinematic control methods have been developed for path-following applications in four-wheeled Mecanum robots [19]. Despite their relevance, these studies are generally centered on Mecanum drives and do not provide a systematic cross-architecture energy comparison that also includes four-omni solutions within the same simulation environment and benchmark trajectory.

2.4. Simulation-Based Modeling in MATLAB/Simulink–Simscape for Reproducible Evaluation

To achieve fair comparisons among locomotion architectures, a unified modeling approach is required to keep mass properties, wheel parameters, contact assumptions, and motion profiles consistent. MATLAB/Simulink–Simscape Multibody R2024a (MathWorks, Natick, MA, USA) has increasingly been used for multibody dynamics and friction-aware modeling, enabling reproducible studies of torque, power, and energy demand. Simulink-based dynamic models for Mecanum-drive platforms have explicitly incorporated frictional forces, thus providing a methodological basis for simulation-based energy analysis [14]. Within the same modeling ecosystem, dynamic models combined with decision-making logic have further demonstrated the suitability of Simulink–Simscape for consistent locomotion evaluation [29]. Moreover, explicit friction integration in Simulink/Simscape is consistent with recent omnidirectional modeling approaches [15,30].

2.5. Synthesis and Positioning of This Work

The reviewed literature shows that recent research on omnidirectional mobile robots has progressed along several complementary directions. First, a substantial body of work addresses robot design, kinematic configuration, and the practical implementation of omni-wheel and Mecanum-wheel platforms for maneuverable robotic systems [1,2,3,4,5]. Second, increasing attention has been given to energy consumption analysis, energy prediction methods, current-based monitoring, and the influence of payload or mission conditions on robot power demand [6,9,10,11,12,13,31]. Third, several studies have investigated wheel–ground interaction, rolling resistance, friction anisotropy, and dynamic effects associated with roller-based wheel structures, highlighting their importance for realistic multibody modeling and for the correct interpretation of platform behavior [7,8,14,15]. Finally, advanced control and motion-planning approaches have been developed to improve trajectory tracking, robustness, and energetic performance in omnidirectional mobile robots [16,18,19,20,21,22,23,24,25,26,27,28,32].

Despite these advances, most published studies remain focused on one locomotion architecture, one methodological layer, or one particular performance objective. In many cases, the literature addresses only kinematic control, only dynamic modeling, only wheel–ground interaction, or only energy-aware trajectory generation, without providing a controlled cross-architecture comparison under identical operating conditions [6,7,8,12,13,14,15,16,18,19,20,21,22,23,24,25,26,27,28,31,32]. As a consequence, the literature still lacks a systematic and fair comparison between two representative four-wheel omnidirectional solutions—namely, a four-omni-wheel platform with 90° roller arrangement and a four-Mecanum-wheel platform with 45° roller arrangement—implemented within the same modeling philosophy and evaluated using the same experimental methodology [6,10,12,14,17,23,29,33].

Table 1. Representative recent studies on omnidirectional locomotion modeling and energy aspects (2020–2025).

Ref. (Title)	Platform/Wheels	Core Method	Energy/Loss Treatment	Main Contribution and Limitation
Practical Model for Energy Consumption Analysis of Omnidirectional Mobile Robot (2021) [6]	Omnidirectional robot (general)	Practical/analytical energy model	Robot-level energy estimation	Strong basis for energy metrics; no unified Simscape benchmark comparison across four-omni and 4-Mecanum.
Energy-Optimal Motion Trajectory of an Omni-Directional Mecanum-Wheeled Robot via Polynomial Functions (2020) [16]	4 Mecanum (45°)	Polynomial trajectory optimization	Energy minimization objective	Relevant for energy-optimal planning; Mecanum-only scope.
A Mecanum Wheel Model Based on Orthotropic Friction with Experimental Validation (2024) [8]	4 Mecanum (45°)	Orthotropic friction + experimental validation	Contact-loss sensitivity	High-fidelity wheel–ground interaction; not a unified energy benchmark across architectures.
Experimental Evaluation of Rolling Resistance in Omnidirectional Wheels Under Quasi-Static Conditions (2025) [7]	Omni wheels (passive rollers)	Quasi-static experiments	Rolling resistance/loss quantification	Provides loss characterization; not a robot-level dynamic energy comparison on identical path.
Simulink Based Dynamic Model for Mecanum Drive Autonomous Mobile Platforms Considering Friction Forces (2024) [14]	4 Mecanum (45°)	MATLAB/Simulink dynamic model	Friction included; energy derivable	Relevant modeling workflow; single architecture (Mecanum); no controlled four-omni vs. four-Mecanum comparison under identical benchmark conditions.
This work (present paper)	Modular robot; 4 omni (90°) vs. 4 Mecanum (45°)	Unified MATLAB/ Simulink–Simscape model + Hall-sensor validation	12 V-based input-power estimate comparison on a 1 m × 1 m square path	Fair cross-architecture comparison; limited to one benchmark trajectory

summarizes representative studies from the recent literature and highlights the positioning of the present work relative to existing contributions. In contrast to previous studies, the present paper does not focus exclusively on one wheel technology or on a single isolated aspect such as control, friction modeling, or energy prediction. Instead, it combines CAD-consistent multibody modeling, identical benchmark conditions, and current-based experimental validation in order to perform a direct and reproducible comparison between two widely used four-wheel omnidirectional locomotion architectures [10,14,17,23,29,33].

The main research novelties of this paper can be summarized as follows:

• A unified MATLAB/Simulink–Simscape Multibody R2024a (MathWorks, Natick, MA, USA) dynamic modeling workflow is established for two four-wheel omnidirectional locomotion architectures implemented on the same modular mobile robot platform [14,17,23,29].
• A controlled comparative analysis is carried out under identical operating conditions, including the same supply voltage, the same test duration, and the same benchmark trajectory, which reduces the influence of external comparison bias and supports a fair evaluation of the two architectures [10,17,33].
• The simulated current demand and the corresponding 12 V-based input-power estimate are experimentally assessed through Hall-effect current sensing integrated into the drive modules, enabling a direct simulation-to-measurement comparison for both locomotion solutions [10].
• The study discusses the influence of roller geometry and wheel–ground interaction on both current demand (and the corresponding electrical input-power estimate) and on model fidelity, thereby providing practical guidance for the selection of locomotion architecture in modular omnidirectional mobile robots [7,8,14,15].

Overall, the present work is positioned at the intersection of dynamic modeling, power-oriented performance evaluation, and experimental validation and addresses a gap that is only partially covered in the recent literature on omnidirectional mobile robotics [6,10,12,14,17,23,28,29].

3. Materials and Methods

3.1. Modular Mobile Platform and Analyzed Locomotion Architectures

This work performs a comparative assessment of current demand and 12 V-based input-power estimate for two four-wheel omnidirectional locomotion architectures implemented on the same modular mobile platform. The objective is to isolate the effect of the locomotion architecture on current demand and corresponding 12 V-based input-power estimate by keeping the chassis structure and the overall modeling framework consistent across cases. The evaluated configurations are: (a) four omni wheels (90° rollers) and (b) four Mecanum wheels (45° rollers). The two configurations and the motor/wheel indexing convention (M1–M4) are shown in Figure 1, which also establishes the reference geometry used later for kinematic modeling and for the comparative analysis of current demand and 12 V-based input-power estimate. The selection of these architectures is aligned with recent studies that discuss the maneuverability–loss trade-off for omni and Mecanum wheels and their deployment in real application contexts [1,33].

3.2. CAD-to-MATLAB/Simulink–Simscape Workflow and Multibody Model Generation

To ensure geometric and inertial consistency across architectures, the platform 3D CAD model was integrated into MATLAB/Simulink–Simscape Multibody R2024a (MathWorks, Natick, MA, USA). The adopted workflow (CAD export → MATLAB import → multibody chain generation) is summarized in Figure 2, while the outcome of the CAD import (rigid bodies, joints, and reference frames) is illustrated in Figure 3.

This workflow supports reproducible multibody modeling, consistent logging of joint forces/torques, and repeatable dynamic simulations. In this study, the comparative assessment is performed using a 12 V-based electrical input-power estimate computed from measured and simulated motor currents under a nominal constant 12 V supply assumption [22].

3.3. Kinematic Modeling and Jacobian Matrices

Kinematic modeling is used in two complementary ways: (i) inverse kinematics to impose the platform motion along the benchmark trajectory and (ii) effort projection through the Jacobian and its transpose to connect platform-level generalized forces with wheel-level torque demand. Jacobian-based formulations are widely used in omnidirectional robot modeling and control, including kinematic control for Mecanum platforms [19] and trajectory tracking methods based on nonlinear optimal control or fuzzy/neural structures [21,22].

Moreover, comparative analyses between Mecanum and other drive solutions from the perspective of dynamic loads motivate unified effort modeling in order to enable meaningful cross-architecture comparisons [10].

The Jacobian matrices used for the two configurations are defined as follows.

3.3.1. Four Omni Wheels (Rollers at 90°)

$${\mathrm{J}}_{4\_omni}=\left[\begin{array}{cccc}{\displaystyle \frac{r}{4}}\mathrm{s}\mathrm{i}\mathrm{n}{\alpha }_1& {\displaystyle \frac{r}{4}}\mathrm{s}\mathrm{i}\mathrm{n}{\alpha }_2& {\displaystyle \frac{r}{4}}\mathrm{s}\mathrm{i}\mathrm{n}{\alpha }_3& {\displaystyle \frac{r}{4}}\mathrm{s}\mathrm{i}\mathrm{n}{\alpha }_4\\ {\displaystyle \frac{r}{4}}\mathrm{c}\mathrm{o}\mathrm{s}{\alpha }_1& {\displaystyle \frac{r}{4}}\mathrm{c}\mathrm{o}\mathrm{s}{\alpha }_2& {\displaystyle \frac{r}{4}}\mathrm{c}\mathrm{o}\mathrm{s}{\alpha }_3& {\displaystyle \frac{r}{4}}\mathrm{c}\mathrm{o}\mathrm{s}{\alpha }_4\\ -{\displaystyle \frac{r}{4}}L& -{\displaystyle \frac{r}{4}}L& -{\displaystyle \frac{r}{4}}L& -{\displaystyle \frac{r}{4}}L\end{array}\right],$$

(1)

$${\mathrm{J}}_{4\_omni}^{*}=\left[\begin{array}{ccc}{\displaystyle \frac{r}{4}}\mathrm{s}\mathrm{i}\mathrm{n}{\alpha }_1& {\displaystyle \frac{r}{4}}\mathrm{c}\mathrm{o}\mathrm{s}{\alpha }_1& -{\displaystyle \frac{r}{4}}L\\ {\displaystyle \frac{r}{4}}\mathrm{s}\mathrm{i}\mathrm{n}{\alpha }_2& {\displaystyle \frac{r}{4}}\mathrm{c}\mathrm{o}\mathrm{s}{\alpha }_2& -{\displaystyle \frac{r}{4}}L\\ {\displaystyle \frac{r}{4}}\mathrm{s}\mathrm{i}\mathrm{n}{\alpha }_3& {\displaystyle \frac{r}{4}}\mathrm{c}\mathrm{o}\mathrm{s}{\alpha }_3& -{\displaystyle \frac{r}{4}}L\\ {\displaystyle \frac{r}{4}}\mathrm{s}\mathrm{i}\mathrm{n}{\alpha }_4& {\displaystyle \frac{r}{4}}\mathrm{c}\mathrm{o}\mathrm{s}{\alpha }_4& -{\displaystyle \frac{r}{4}}L\end{array}\right].$$

(2)

3.3.2. Four Mecanum Wheels (Rollers at 45°)

$${\mathrm{J}}_{mecanum}=\left[\begin{array}{ccc}{\displaystyle \frac{r}{4}}& -{\displaystyle \frac{r}{4}}& -{\displaystyle \frac{r}{2}}\left(L_x+L_y\right)\\ {\displaystyle \frac{r}{4}}& {\displaystyle \frac{r}{4}}& {\displaystyle \frac{r}{2}}\left(L_x+L_y\right)\\ {\displaystyle \frac{r}{4}}& {\displaystyle \frac{r}{4}}& -{\displaystyle \frac{r}{2}}\left(L_x+L_y\right)\\ {\displaystyle \frac{r}{4}}& -{\displaystyle \frac{r}{4}}& {\displaystyle \frac{r}{2}}\left(L_x+L_y\right)\end{array}\right],$$

(3)

$${\mathrm{J}}_{mecanum}^{*}=\left[\begin{array}{cccc}{\displaystyle \frac{r}{4}}& {\displaystyle \frac{r}{4}}& {\displaystyle \frac{r}{4}}& {\displaystyle \frac{r}{4}}\\ -{\displaystyle \frac{r}{4}}& {\displaystyle \frac{r}{4}}& {\displaystyle \frac{r}{4}}& -{\displaystyle \frac{r}{4}}\\ -{\displaystyle \frac{r}{2}}\left(L_x+L_y\right)& {\displaystyle \frac{r}{2}}\left(L_x+L_y\right)& {\displaystyle \frac{r}{2}}\left(L_x+L_y\right)& -{\displaystyle \frac{r}{2}}\left(L_x+L_y\right)\end{array}\right].$$

(4)

3.4. Simulink–Simscape Dynamic Models and Loss Inclusion

Dynamic models for the two configurations are implemented in Simulink–Simscape–Multibody and include: the imported multibody chain, wheel actuation via “Revolute” joints, sensing/logging of resistive torques, and dedicated blocks for friction-related losses. The model structures for the four-omni-wheel and Mecanum-wheel configurations are shown in Figure 4 and Figure 5. This modeling strategy supports friction-aware dynamic evaluation and provides motor-level signals that can be compared to experimental current measurements for current-demand and 12 V-based input-power assessment. For Mecanum drives, the impact of payload and disturbances motivates analyzing torque and power under varying load conditions [9,23]. In addition, anisotropic contact losses for Mecanum wheels have been modeled and validated experimentally [8], while omni-wheel losses can be parameterized via rolling resistance characterization [7]. Contact and friction assumptions used in the present model should be stated explicitly. In the adopted MATLAB/Simulink–Simscape Multibody R2024a (MathWorks, Natick, MA, USA) framework, the model is used as a friction-aware comparative multibody model under identical assumptions for both locomotion architectures so that the observed differences can be interpreted within a common and reproducible simulation structure. However, for the present study, the Mecanum-wheel contact is not represented through a fully calibrated orthotropic friction law resolved at passive-roller level. Instead, the model captures the global actuation response within a unified simulation framework, without explicitly resolving the detailed local slip kinematics, roller-level contact forces, or passive-roller dynamics for each individual roller. This means that the present formulation is primarily intended to support a controlled cross-architecture comparison and a simulation-to-experiment assessment under the selected benchmark conditions rather than to provide a high-fidelity predictive contact model for all Mecanum-wheel operating regimes. In particular, the local anisotropic effects associated with the 45° roller arrangement, including direction-dependent slip and additional losses at roller level, are only represented indirectly through the adopted friction-aware multibody structure. As a consequence, the present model is expected to reproduce the global dynamic tendencies of the platform more reliably than the full local contact mechanics of each wheel–ground interaction point. This limitation is especially relevant for the four-Mecanum configuration, where the simulation-to-experiment deviation is larger, and it indicates that further refinement through parameter identification and more detailed friction/contact modeling would be beneficial in future work. Nevertheless, within the scope of the present study, the adopted model remains suitable for a fair and consistent benchmark-based comparison between the two interchangeable locomotion architectures.

3.5. Relations for Power and Energy Evaluation

To make the transition from the dynamic model outputs to the adopted input-power assessment explicit, the electromechanical relations used in this study are summarized in this subsection. In the adopted modeling framework, the generalized forces associated with the platform motion are projected into wheel-level effort through the transpose Jacobian, while the Simulink–Simscape dynamic model provides actuator-level variables that can be related to the experimentally measured motor currents [10,23,29]. Since both simulation and experiment are evaluated under the same nominal 12 V supply assumption, the resulting metric is interpreted in this study as a 12 V-based electrical input-power estimate, suitable for controlled comparative evaluation under identical benchmark conditions.

The wheel torque vector is obtained from the generalized force vector acting at platform level according to:

$${\tau }_w={\mathrm{J}}^T\mathrm{F}$$

(5)

where $\mathrm{F}$ is the vector of generalized forces associated with the imposed platform motion, $\mathrm{J}$ is the Jacobian matrix corresponding to the selected locomotion architecture, and ${\tau }_w$ is the wheel torque vector. This relation is consistent with the effort projection approach introduced in Section 3.3.

For each wheel–motor unit, the instantaneous mechanical power can be expressed as:

$$P_{\mathrm{m}\mathrm{e}\mathrm{c}\mathrm{h},i}\left(t\right)={\tau }_i\left(t\right)\hspace{0.17em}{\omega }_i\left(t\right)$$

(6)

where ${\tau }_i\left(t\right)$ is the instantaneous actuator torque, and ${\omega }_i\left(t\right)$ is the angular speed of motor/wheel $i$.

Assuming DC motor actuation, the motor torque is proportional to the current drawn by the corresponding actuator. Thus, the torque–current relation can be written as:

$${\tau }_i\left(t\right)=K_t{I}_i\left(t\right)$$

(7)

or equivalently:

$$I_i\left(t\right)={\displaystyle \frac{{\tau }_i\left(t\right)}{K_t}}$$

(8)

where $K_t$ is the motor torque constant, and $I_i\left(t\right)$ is the motor current. This relation provides the missing link between the torque demand obtained from the dynamic model and the current-based experimental validation approach adopted in this paper [10].

Because both the simulations and the experiments are evaluated under a nominal constant 12 V supply, the instantaneous electrical input-power estimate of actuator i is expressed as:

$$P_{\mathrm{e}\mathrm{l},i}\left(t\right)=U\hspace{0.17em}{I}_i\left(t\right)$$

(9)

where $U=12 \mathrm{V}$ is the supply voltage. The total 12 V-based electrical input-power estimate of the robot is then obtained by summing the contribution of the four drive motors:

$$P_{\mathrm{t}\mathrm{o}\mathrm{t}}\left(t\right)=\sum _{i=1}^4{P}_{\mathrm{e}\mathrm{l},i}\left(t\right)=U\sum _{i=1}^4{I}_i\left(t\right)$$

(10)

For completeness, the corresponding 12 V-based electrical input-energy estimate over a trajectory of duration T is defined as:

$$E={\int }_0^T{P}_{\mathrm{t}\mathrm{o}\mathrm{t}}\left(t\right)\hspace{0.17em}dt$$

(11)

In the present study, the comparative evaluation is mainly based on the total instantaneous input-power estimate, the peak input-power estimate, and the average input-power estimate over the benchmark trajectory. Nevertheless, Equations (5)–(11) clarify the electromechanical chain that links the generalized-force formulation and the dynamic model outputs to the measured motor currents and, consequently, to the adopted 12 V-based electrical input-power estimate used for the comparison between the two locomotion architectures [10,23,29]. This metric is appropriate for controlled comparative assessment under identical supply assumptions, but it should not be interpreted as a complete decomposition of all electrical, mechanical, and drivetrain losses within the actuation system.

3.6. Benchmark Trajectory and 12 V-Based Input-Power Estimate Comparison

The comparative analysis of current demand and 12 V-based input-power estimate is performed on a standardized reference benchmark consisting of a 1 m side-length square path. This benchmark was selected because it combines repeated straight-line motion with successive 90° direction changes, thus generating both quasi-steady motion intervals and transient torque and input-power estimate peaks associated with acceleration, deceleration, and reorientation. Such a trajectory is suitable for a controlled cross-architecture comparison because it imposes the same motion logic, the same supply conditions, and the same test duration for both locomotion solutions.

Figure 6 presents the benchmark trajectory used in the present study. The purpose of this benchmark is not to represent the full operating envelope of omnidirectional mobile robots but, rather, to provide a reproducible and identical reference maneuver for comparing the current demand and corresponding 12 V-based input-power estimate of the four-omni and four-Mecanum configurations under the same experimental and simulation framework. Similar square-path benchmark logic has also been used in related current- and energy-oriented studies on mobile robots [10,17].

4. Results

This section presents the results obtained for two four-wheel omnidirectional locomotion architectures (omni wheels with 90° rollers and Mecanum wheels with 45° rollers) evaluated on the same benchmark (1 m square trajectory, 4 m total path length) under identical supply voltage (12 V) and test duration. The following results are reported: (i) the physical robot configurations and the main constructive characteristics, (ii) dynamic modeling and MATLAB/Simulink–Simscape Multibody R2024a (MathWorks, Natick, MA, USA) simulation outcomes, (iii) experimental current measurements acquired using Hall-effect sensors, and (iv) simulation-to-experiment comparisons for current and 12 V-based input-power estimate, followed by a direct benchmark-based comparison between the two configurations.

4.1. Robot Configurations and Main Characteristics

This subsection documents the two physical configurations investigated and summarizes the parameters required to ensure fair comparability (dimensions, mass, supply voltage, wheel type, and acquisition hardware). Both configurations are implemented on the same modular platform using the same control and data acquisition core; the primary difference is the wheel architecture (omni vs. Mecanum) and the roller geometry. The four-omni-wheel configuration (90° rollers) is shown in Figure 7. The image captures the real platform during the experimental runs, highlighting the integration of the wheel modules and the overall mechanical layout used for the benchmark trajectory.

The four-Mecanum-wheel configuration (45° rollers) is shown in Figure 8. The figure presents the real modular platform during the experimental campaign, highlighting the replacement of the wheel modules with Mecanum wheels featuring 45° inclined rollers while preserving the same chassis, drive layout, and onboard hardware. This ensures that the comparison between architectures is performed under consistent conditions, i.e., identical mechanical frame, motor/driver arrangement and sensing and control hardware and the same benchmark setup. Consequently, the observed differences in current demand and corresponding 12 V-based input-power estimate reported in Section 4 can be primarily attributed to the wheel–ground interaction characteristics associated with the Mecanum roller geometry rather than to changes in the underlying platform.

The main constructive and operational characteristics of the two configurations are summarized in

Table 2. Summary of the main technical characteristics for the two configurations (four-omni vs. four-Mecanum).

Parameter	Four-Omni (90° Rollers)	Four-Mecanum (45° Rollers)
Dimensions	Ground footprint approximated by a circle of 300 mm radius	493 × 454 × 210 mm (L × W × H)
Robot mass	23 kg	23 kg
Maximum payload	~10 kg	~10 kg
Maximum speed	0.5 m/s	0.5 m/s
Maximum acceleration	1 m/s2	1 m/s2
Motor supply voltage	12 V	12 V
Maximum motor torque	Max. 7.3 Nm	Max. 7.3 Nm
Wheel diameter	100 mm	100 mm
Wheel type	Omni wheel, 90° rollers	Mecanum wheel, 45° rollers
Current sensing	ACS712-5A Hall-effect current sensors (Allegro MicroSystems, Worcester, MA, USA)	ACS712-5A Hall-effect current sensors (Allegro MicroSystems, Worcester, MA, USA)
Control hardware	Raspberry Pi 4 Model B (Raspberry Pi Ltd., Cambridge, UK), Arduino Mega 2560 Rev3 (Arduino, Monza, Italy)	Raspberry Pi 4 Model B (Raspberry Pi Ltd., Cambridge, UK), Arduino Mega 2560 Rev3 (Arduino, Monza, Italy)
Software	ROS integration capability	ROS integration capability

, compiled from the platform specifications of the omni-wheel and Mecanum-wheel setups. The table reports the key parameters required for a fair comparison—such as overall dimensions, robot mass and payload capacity, wheel diameter and type, supply voltage, and maximum speed and acceleration, as well as the sensing and control hardware used for current acquisition and robot operation.

4.2. Dynamic Modeling, Simulation, and Experimental Validation

Dynamic modeling was implemented in MATLAB/Simulink–Simscape Multibody R2024a (MathWorks, Natick, MA, USA) for both configurations using the same methodology to estimate the motor currents along the imposed trajectory. Experimental validation was performed by measuring the current drawn by each motor using integrated Hall-effect sensors. For each configuration, the following are presented: the block diagram of the simulation/acquisition workflow, simulated current, measured current, simulation-to-experiment current comparison, and simulation-to-experiment comparison of the 12 V-based input-power estimate.

4.2.1. Four-Omni (90° Rollers): Simulation, Measurement, and Comparison

The simulation and acquisition workflow for the four-omni configuration is shown in Figure 9. The block diagram outlines the sequence from trajectory execution and dynamic model evaluation to Hall-based current acquisition and direct simulation-to-experiment comparison.

The simulated motor current profiles for the four-omni configuration are shown in Figure 10. The figure presents the time evolution of the currents for the four motors over the square trajectory, reflecting straight segments and direction changes.

The measured motor current profiles acquired with Hall sensors for the same configuration are shown in Figure 11. The data are recorded under the same supply voltage and test duration as used in simulation to support a consistent comparison.

A direct comparison between simulated and measured currents for the four-omni configuration is presented in Figure 12, enabling assessment of global agreement and local deviations.

Figure 13 compares the simulated and measured input-power estimate for the four-omni configuration at U = 12 V.

The agreement between simulated and measured motor-current waveforms for the four-omni configuration is summarized in

Table 3. Waveform-level agreement indicators (simulated vs. measured motor currents) for the four-omni configuration (90° rollers).

Motor	RMSE (A)	MAE (A)	Pearson Correlation Coefficient (-)
M1	0.266	0.170	0.912
M2	0.245	0.172	0.930
M3	0.239	0.132	0.911
M4	0.245	0.126	0.930

using waveform-level comparison indicators, namely the root mean square error (RMSE), the mean absolute error (MAE), and the Pearson correlation coefficient. These indicators were selected because the current signals are bidirectional and may include near-zero central values, for which median-based relative errors can become numerically unstable and less informative for model-assessment purposes.

The waveform-level indicators show a good qualitative agreement between simulation and experiment for the four-omni configuration, with high correlation values for all four motors and moderate error levels consistent with the adopted simplified friction-aware multibody model.

4.2.2. Four-Mecanum (45° Rollers): Simulation, Measurement, and Comparison

The simulation and acquisition workflow for the four-Mecanum configuration is shown in Figure 14. The block diagram follows the same processing chain as for the four-omni case, ensuring methodological consistency (simulation → Hall-based acquisition → comparison).

The simulated motor current profiles for the four-Mecanum configuration are shown in Figure 15, computed for the same 1 m × 1 m square benchmark (4 m total path length) under the identical test duration and a 12 V supply. The figure reports the time evolution of the simulated currents for the four drive motors (M1–M4) during the complete trajectory execution, capturing both straight-line segments and corner transitions. These simulated signals are used as the reference model output for the subsequent validation against the Hall-sensor measurements presented in the following subsection.

The measured motor current profiles acquired with Hall sensors for the four-Mecanum configuration are shown in Figure 16, recorded under the same supply voltage (12 V) and identical test duration as used for simulation.

A direct comparison between simulated and measured currents for the four-Mecanum configuration is presented in Figure 17, supporting the assessment of agreement between model outputs and experimental signals.

The simulated versus measured input-power estimate comparison for the four-Mecanum configuration is shown in Figure 18, where the 12 V-based input-power estimate is computed under the nominal constant supply assumption U = 12 V.

Similarly,

Table 4. Waveform-level agreement indicators (simulated vs. measured motor currents) for the four-Mecanum configuration (45° rollers).

Motor	RMSE (A)	MAE (A)	Pearson Correlation Coefficient (-)
M1	0.186	0.117	0.954
M2	0.167	0.100	0.927
M3	0.180	0.097	0.907
M4	0.256	0.119	0.867

summarizes the agreement between simulated and measured motor-current waveforms for the four-Mecanum configuration using RMSE, MAE, and the Pearson correlation coefficient. These indicators provide a more robust description of simulation-to-experiment agreement for bidirectional current signals than median-based relative errors, especially in cases where the current waveform contains values close to zero.

The waveform-level indicators show a generally good qualitative agreement between simulation and experiment for the four-Mecanum configuration, although the error levels remain consistent with the more complex wheel–ground interaction associated with the 45° roller geometry.

4.3. Comparative Assessment of the 12 V-Based Input-Power Estimate Between the Two Configurations

To support a benchmark-based comparison,

Table 5. Current balance and corresponding 12 V-based input-power estimate (simulated vs. measured) for the four-omni configuration (90° rollers) under the nominal supply assumption U = 12 V.

Quantity	M1	M2	M3	M4	Total
Measured current (A)	0.58	0.50	0.48	0.60	2.16
Simulated current (A)	0.54	0.59	0.38	0.57	2.08
Current error (%)	5.92	−18.99	20.91	4.90	3.70
12 V-based input-power estimate from measurements (W)	6.92	5.95	5.74	7.14	25.75
12 V-based input-power estimate from simulation (W)	6.51	7.08	4.54	6.79	24.92
Input-power estimate error (%)	5.92	−18.99	20.91	4.90	3.70

and

Table 6. Current balance and corresponding 12 V-based input-power estimate (simulated vs. measured) for the four-Mecanum configuration (45° rollers) under the nominal supply assumption U = 12 V.

Quantity	M1	M2	M3	M4	Total
Measured current (A)	0.41	0.36	0.33	0.38	1.48
Simulated current (A)	0.30	0.31	0.27	0.30	1.18
Current error (%)	26.77	15.47	19.45	23.21	20.27
12 V-based input-power estimate from measurements (W)	4.93	4.33	4.01	4.61	17.88
12 V-based input-power estimate from simulation (W)	3.61	3.66	3.23	3.54	14.04
Input-power estimate error (%)	26.77	15.47	19.45	23.21	21.42

report the current balance and the corresponding 12 V-based input-power estimate for each motor and for the full robot, in both simulated and measured regimes, for the four-omni and four-Mecanum configurations. For the four-omni configuration, the current balance and the corresponding 12 V-based input-power estimate are provided in Table 5.

For the four-Mecanum configuration, the current balance and the corresponding 12 V-based input-power estimate are provided in Table 6.

Based on the total values reported in Table 5 and Table 6,

Table 7. Direct comparison of total 12 V-based input-power estimate between the four-omni and four-Mecanum configurations under identical benchmark conditions.

Configuration	Total Measured Input-Power Estimate (W)	Total Simulated Input-Power Estimate (W)	Robot-Level Deviation (%)
Four-omni (90° rollers)	25.75	24.92	3.70
Four-Mecanum (45° rollers)	17.88	14.04	21.42

provides a direct comparison of the total measured and simulated 12 V-based input-power estimate for the two configurations, together with the corresponding simulation-to-measurement deviation at robot level.

5. Discussion

This section discusses the main implications of the results presented in Section 4, focusing on: (i) the differences between the two locomotion architectures (four-omni with 90° rollers and four-Mecanum with 45° rollers) in terms of measured and simulated current demand and 12 V-based input-power estimate on the square benchmark and (ii) the agreement between the MATLAB/Simulink–Simscape Multibody R2024a (MathWorks, Natick, MA, USA) dynamic model and the experimental measurements obtained using Hall-effect current sensors. The discussion is limited to interpreting the reported figures and tables under the adopted benchmark and nominal-voltage assumptions, without introducing new results.

5.1. Interpretation of the Measured 12 V-Based Input-Power Estimate on the Square Benchmark

The total comparison reported in Table 7 shows that for the considered 1 m × 1 m square trajectory (4 m total path length), under identical nominal supply voltage (12 V) and identical test duration, the four-Mecanum configuration exhibits a lower measured 12 V-based input-power estimate than the four-omni configuration. This trend is also supported by the motor-level current and corresponding input-power balance reported in Table 5 and Table 6, where the measured currents, and therefore the resulting 12 V-based input-power estimates, are lower for the four-Mecanum case. A technically plausible interpretation is related to how each architecture distributes actuation effort during the benchmark execution. The square trajectory contains repeated straight segments and corner transitions (direction changes), which can impose different loading patterns depending on the wheel type and the associated wheel–ground constraints. Under the tested conditions, the four-Mecanum architecture may lead to a motor operating regime associated with lower average current demand, resulting in a lower 12 V-based input-power estimate.

However, a lower measured 12 V-based input-power estimate does not necessarily imply higher locomotion efficiency. For omnidirectional wheels, especially Mecanum wheels, the wheel–ground interaction can be significantly influenced by micro-slip, directional friction effects, roller losses, and local compliance. In such cases, a reduction in measured current can also reflect reduced effective traction forces due to slip rather than purely improved energetic performance. Therefore, the present comparison should be interpreted as a benchmark-specific comparison of measured current demand and nominal-voltage input-power estimate, rather than as a generalized efficiency ranking of the two locomotion architectures. A trajectory-tracking quality metric would help distinguish lower input demand caused by improved energetic behavior from lower input demand associated with increased slip and is therefore recommended for future studies.

5.2. Simulation-to-Experiment Agreement and Dynamic Model Fidelity

The results indicate different levels of simulation-to-experiment agreement for the two architectures. At robot level, Table 7 shows that the four-omni configuration achieves a small deviation between simulated and measured total input-power estimate, suggesting that the adopted model structure captures the dominant effects governing electrical loading for this configuration on the selected benchmark. In addition, the current comparison in Figure 12 shows a generally good overlap between simulated and measured signals.

In contrast, the four-Mecanum configuration exhibits a substantially larger deviation between simulated and measured total input-power estimate (Table 7), and Figure 17 indicates a more pronounced mismatch between the simulated and measured current profiles. This behavior is consistent with the higher complexity of Mecanum wheel–ground interaction, where longitudinal and lateral force components are coupled by the 45° roller geometry. From a modeling perspective, this interaction can be highly sensitive to factors such as directional friction coefficients, roller rolling resistance, additional mechanical losses, and the specific surface conditions during testing.

Beyond contact modeling, discrepancies can also be influenced by practical measurement-chain effects, including Hall sensor offset and bandwidth limitations, PWM-related ripple from motor drivers, and signal conditioning/filtering. These effects can become more visible when the current levels are lower or when transients are more significant, increasing the relative deviation. Overall, the findings indicate that the current model is more robust for a four-omni configuration in its present form, whereas a four-Mecanum configuration requires additional calibration and/or enhanced loss and contact modeling to achieve comparable predictive accuracy.

At waveform level, the agreement between simulated and measured motor-current signals remains generally good for both configurations, as indicated by the RMSE, MAE, and correlation values reported in Table 3 and Table 4. However, robot-level agreement in terms of total 12 V-based input-power estimate remains more favorable for the four-omni configuration, which suggests that local waveform agreement alone does not fully capture the cumulative influence of wheel–ground interaction and modeling simplifications at platform level.

5.3. Implications for Locomotion Architecture Selection in a Modular Robot

Given the modular nature of the platform and the ability to interchange wheel modules, the results suggest two complementary, application-driven conclusions:

• From the measured input-power-estimate perspective: the four-Mecanum configuration yields a lower measured total input-power estimate than the four-omni configuration on the evaluated benchmark (Table 7). For applications where limiting instantaneous electrical input-power estimate is critical (e.g., battery autonomy, power budgeting, or thermal constraints at the driver level), this result favors the four-Mecanum architecture under the tested conditions.
• From the model-based design and predictability perspective: the four-omni configuration provides a closer simulation-to-experiment match at robot level (Table 7), which is valuable for model-based development tasks such as parameter tuning, scenario-based virtual testing, and controller design prior to implementation.

Therefore, architecture selection should not rely on a single metric but, rather, on the target objective. If the priority is reducing measured input-power estimate on the considered motion profile, four-Mecanum configuration is advantageous; if the priority is a reliable predictive model that supports rapid design iteration and robust validation, four-omni configuration offers benefits with the current modeling approach. Accordingly, the present conclusions should be interpreted with reference to the analyzed benchmark trajectory and should not yet be generalized to all omnidirectional motion regimes without additional trajectory-level validation.

5.4. Limitations and Immediate Directions for Improvement

The scope of the present study is defined by several limitations that also indicate immediate directions for future development:

• Single benchmark trajectory: the selected 1 m square path provides a controlled and reproducible reference case for comparative evaluation, since it combines straight segments and repeated 90° direction changes under identical operating conditions. However, it does not cover the full motion envelope of omnidirectional mobile robots. Additional trajectories, such as diagonal motion, rotation-in-place, curved paths, and combined translation-rotation maneuvers, should be included in future work to assess the robustness and generality of the observed current demand and input-power-estimate trends.
• Input-power estimate computed under nominal voltage assumption: the 12 V-based input-power estimate was computed using current signals and a nominal constant supply voltage (12 V). Measuring the actual voltage at the driver input during motion would capture voltage sag and improve power-estimation accuracy.
• Contact modeling for Mecanum wheels: the simulation-experiment mismatch suggests extending the model to include directional friction, roller resistance, and drivetrain/driver losses, supported by parameter identification on the test surface.
• Absence of a tracking-accuracy metric: adding an objective trajectory-tracking metric, such as external localization or validated odometry, would clarify the relationship between input-power estimate and kinematic performance in the presence of slip. These limitations do not alter the reported results for the considered benchmark, but they define a clear path for extending and consolidating the comparative evaluation in future work.

6. Conclusions and Future Research Directions

This paper presented a comparative modeling and experimental validation framework for a modular four-wheel omnidirectional mobile robot, with the main objective of comparing two locomotion architectures—four-omni (omni wheels with 90° rollers) and four-Mecanum (Mecanum wheels with 45° rollers)—using a 12 V-based electrical input-power estimate as the primary comparative metric. Both configurations were evaluated under identical conditions, using the same supply voltage (12 V) and the same test duration, on a standardized reference benchmark consisting of a 1 m × 1 m square trajectory (4 m total path length). Experimental validation was performed by measuring motor currents using Hall-effect sensors, while dynamic simulation was implemented in MATLAB/Simulink–Simscape Multibody R2024a (MathWorks, Natick, MA, USA).

The experimental results indicate that for the considered benchmark and under the adopted nominal-voltage and test-duration assumptions, the four-Mecanum configuration exhibited a lower measured total 12 V-based input-power estimate than the four-omni configuration (17.88 W versus 25.75 W, as reported in Table 7). Regarding simulation-to-experiment agreement at robot level, the four-omni configuration exhibited a smaller deviation between simulated and measured total input-power estimate (3.70%), whereas the deviation was larger for the four-Mecanum configuration (21.42%, Table 7). These findings suggest that for the analyzed square-path benchmark and under the current modeling assumptions, the four-Mecanum architecture is associated with a lower measured nominal-voltage input demand, while the four-omni architecture is reproduced with higher fidelity by the adopted dynamic model in its present form.

These findings should be interpreted as benchmark-specific comparative results and should not be generalized to all omnidirectional operating regimes without additional trajectory-level validation, direct voltage measurement during motion, and independent tracking-accuracy assessment.

To strengthen the generality of the conclusions and improve model fidelity—especially for Mecanum-wheel locomotion—the following research directions are recommended:

• Expanded benchmark set: evaluate additional trajectories and motion primitives (straight-line motion at multiple speeds, diagonal motion, rotation-in-place, and combined translation–rotation) to cover a broader operating envelope beyond the square path.
• Contact and friction parameter identification: identify surface-specific friction/contact parameters and include anisotropic traction effects, which are particularly relevant for Mecanum wheels due to coupled longitudinal–lateral interactions and micro-slip phenomena.
• Enhanced loss modeling: explicitly model drivetrain, roller, and motor driver losses (including PWM-related effects and driver efficiency) to reduce simulation–experiment discrepancies.
• Voltage monitoring for improved input-power estimation: measure the actual voltage at the motor driver input during motion to capture voltage sag and transient effects, improving the accuracy of the computed electrical input-power estimate.
• Trajectory-tracking metric as a companion indicator: complement the input-power-estimate-based analysis with an objective tracking-accuracy metric (e.g., external localization or validated odometry) to correlate current demand and input-power estimate with kinematic performance, particularly under slip conditions.
• Repeatability under varied conditions: repeat experiments across different payloads, floor materials, and speed regimes to quantify robustness and determine each architecture’s sensitivity to operating conditions.
• Model-based control integration: use the validated model as a basis for advanced control and energy-aware strategies (e.g., adaptive speed profiling, predictive control, or adaptive mode selection) and evaluate their impact on current demand, input-power estimate, and tracking performance.

Implementing these directions would extend the present study by broadening the applicability of the comparative conclusions and by enabling a more accurate predictive model for design and control under real operating conditions.