Review Reports - Bridging Regimes: A State-Dependent Blending Methodology for Parsimonious and Robust Heavy Vehicle Dynamics Modeling

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Study presents aclear and practical methodology enabling low-order grey-box heavy-vehicle models to generalise across regimes through state-dependent parameter blending grounded in real experimental data. Two physically interpretable parameter sets derived from Constant Radius (steady-state) and Increasing Steer (transient) manoeuvres form “expert poles” that are dynamically interpolated. The optimisation process identifies steering angle as robust scheduling variable, outperforming speed- and derivative-based heuristics. Dynamic Model (“Angle-Only”), tested against a globally tuned Static Model and a common lateral-acceleration Heuristic Model, improves lateral-acceleration accuracy and stability while clearly showing where heuristic blending fails in transient manoeuvres.

Descrined approach is restricted due to two-regime calibration, which does not capture essential variations arising from load state, friction changes, braking-in-turn, or high-speed combined-slip behaviour. The angle-only scheduling variable, although optimal for the tested scenarios, may not reliably indicate regime transitions driven by factors other than steering input limiting its broader applicability.

Authors should consider below improvements of article:

a) adopt a composite scheduling variable in formulation where steering angle remains dominant but is complemented by a slow state like speed or GG-diagram index
b) clarify how chosen steering thresholds in Equation (7) were selected and show( using existing data) that method is not sensitive to these specific values
c) improve clarity by adding important details of optimisation procedure like algorithm, convergence, cost-function weights

d)provide some formal stability explanation to support behaviour shown in Figure 4

e) discuss how the method could be extended from two regimes to a multi-pole structure like additional poles for load changes or friction

Authors should consider extending Introduction for wider readers perspective about real-world operational conditions including challenges underlined in recent studies on vehicle deployment in unstructured terrain (https://doi.org/10.24425/bpasts.2025.153431) and influence of unpaved-road conditions on heavy-vehicle dynamics (https://doi.org/10.21203/rs.3.rs-5010781/v1).

Additionaly some plots like Fig. 1 could use zoomed regions to highlight transient deviations.

Author Response

Comment 1: Authors should consider adopting a composite scheduling variable in formulation where steering angle remains dominant but is complemented by a slow state like speed or GG-diagram index.

Response 1: Thank you for this insightful suggestion. In the early stages of our research, we indeed hypothesized that a composite scheduling variable—incorporating longitudinal speed (u) and steering rate () alongside steering angle (δ_sw)—would provide better regime detection. To test this, we developed a specific optimization script described in Section 3.1. This script performs an optimization based on minimizing the Root Mean Square Error (RMSE) between the calculated blending weight and the definitive target regimes (0 for Steady-State/CR, 1 for Transient/IS).

As presented in Table 2, the optimization results are definitive: the algorithm converged to a solution where the weights for speed and steering rate were driven to zero, leaving the steering angle (δ_sw) as the sole effective scheduling variable. Although we do not explicitly filter for noise in the initial cost function, the optimization process naturally eliminated the derivative (rate) and speed terms, likely because they did not provide a stable correlation with the regime targets for this specific heavy vehicle dataset compared to the robust signal of the steering angle. However, we agree that for broader operational conditions (e.g., varying loads or friction), a composite variable would be necessary. We have added a discussion on this in the relevant section of the revised manuscript.

Comment 2: Clarify how chosen steering thresholds in Equation (7) were selected and show (using existing data) that method is not sensitive to these specific values.

Response 2: The selection of the threshold values in Equation (7) was driven by the specific physical constraints of the test vehicle. The vehicle used in this study is a city bus with a steering system capable of rotating exactly 1110 degrees in both directions. To ensure the blending function is physically realistic and covers the vehicle's entire operational envelope without premature saturation, = 1110 degrees was selected to match this physical limit. was selected near zero to capture the onset of maneuvering. Since these values were anchored to the hard physical limits of the steering actuator rather than arbitrary tuning parameters, the method provides a consistent mapping of the available steering range to the blending weight. We have clarified this physical rationale in Section 2.3 of the revised manuscript.

Comment 3: Improve clarity by adding important details of optimization procedure like algorithm, convergence, cost-function weights.

Response 3: We have revised Section 2 to explicitly detail the two-stage optimization framework employed in this study, addressing both parameter calibration and blending architecture selection:

Stage 1: Expert Parameter Calibration (The Poles):

To define the PARAMS_CR and PARAMS_IS sets, we used a hybrid optimization approach to avoid local minima. A Genetic Algorithm (GA) (Population: 60, Generations: 100) performed the global search, followed by a Pattern Search (PS) algorithm for local refinement.

The cost function was a weighted RMSE sum:

where weights were empirically tuned to lateral stability (), yaw tracking (), and ride comfort () with penalty terms for physical violations (e.g., wheel lift).

Stage 2: Blending Variable Optimization:

To select the scheduling variable (Section 2.3), we formulated a secondary optimization problem minimizing the error between the calculated blending weight and the target regime (0 for Steady-State and 1 for Transient). Using fmincon (Sequential Quadratic Programming - SQP algorithm), we found that the weights for speed and steering rate converged to zero, isolating the Steering Angle as the robust optimal variable.

We have added these algorithmic details and weight definitions to the "Material and Methods" section to ensure full reproducibility.

Reviewer Comment 4: Provide some formal stability explanation to support behaviour shown in Figure 4.

Response 4: The stability difference observed in Figure 4 (newly Figure 6) is directly attributable to the physical properties of the "Expert Poles" identified in our dual calibration (Table 3). The Static Model relies on a single parameter set that approximates the steady-state (CR) behavior, which is characterized by lower damping (). In contrast, the Dynamic Model adapts by shifting towards the transient (IS) parameter set during the step-steer maneuver. As shown in Table 3, the IS pole possesses significantly higher effective damping () and a lower peak force factor (reduces from 6.37 to 3.77). This adaptation allows the Dynamic Model to dissipate the sudden energy input of the step steer, preventing the oscillatory instability exhibited by the under-damped Static Model. We have added this physical explanation to the results section.

Reviewer Comment 5: Discuss how the method could be extended from two regimes to a multi-pole structure like additional poles for load changes or friction.

Response 5: While this study focused on validating the methodology using two fundamental regimes (Steady-State vs. Transient), the proposed blending architecture is mathematically extensible to an N-pole structure. The current linear interpolation, , can be generalized to a weighted summation form: , where represents additional expert parameter sets (poles) calibrated for different conditions, such as varying laden states or friction coefficients (). We have expanded the Discussion section to outline this theoretical framework for multi-pole extension.

Reviewer Comment 6: Authors should consider extending Introduction for wider readers perspective about real-world operational conditions including challenges underlined in recent studies on vehicle deployment in unstructured terrain and influence of unpaved-road conditions on heavy-vehicle dynamics.

Response 6: We agree that framing the study within the broader context of real-world operational challenges strengthens the paper. We have expanded the Introduction to discuss the complexity of heavy vehicle dynamics in diverse environments, specifically incorporating the suggested studies by Nowakowski & Kosiuczenko regarding unstructured terrain and Dantas & Faria regarding unpaved road conditions. These references highlight that real-world operational envelopes are fraught with environmental disturbances that challenge static parameterizations, supporting our argument for adaptive modeling.

Reviewer Comment 7: Additionally some plots like Fig. 1 (newly Figure 3) could use zoomed regions to highlight transient deviations.

Response 7: To enhance the readability of the figures and better highlight transient behaviors, we have revised the layout of the benchmarking plots. The original figures (Fig. 3-6) comprised 12 subplots. To optimize the graphical space, we omitted the input data plots (Vehicle Speed and Steering Angle) and the derived safety metrics (Understeer/Oversteer Tendency and Slipping Risk), thereby reducing the total number of subplots to eight. Furthermore, legend labels have been abbreviated to declutter the plot area and ensure that the key dynamic responses are clearly legible.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The revised manuscript presents a methodically refined and practically relevant approach to vehicle dynamics modeling, the results of which have direct application in the design and validation of chassis and steering actuator control systems. The authors clearly demonstrate that the proper selection of a blending variable significantly improves the stability and accuracy of models used in control, which is crucial for the safe operation of modern ADAS systems. The work bridges the existing gap between complex data models and lightweight physical models, offering an approach that is simultaneously interpretable, computationally efficient, and compatible with actuator systems used in heavy vehicles. Studies involving boundary maneuvers demonstrate the real impact of incorrect or correct modeling on actuator performance, highlighting the technical and practical value of the article for the Actuators community. With a few editorial corrections and strengthened links to the actuator design context, the text will be even more readable and more firmly embedded in the journal's scope. Therefore, in the reviewer's opinion, the article constitutes a valuable contribution and, with modifications, should be accepted for publication in MDPI Actuators. The reviewer believes the following corrections and additions should be made:
1) It is worthwhile to add 2-3 literature sources from the LPV and SDRE areas for heavy vehicles, as this will allow for a stronger contextualization of the problem within the context of contemporary state-dependent control methods. Publications in this field demonstrate that, despite the widespread use of adaptive architectures, clear rules for switching variable selection are still lacking. Supplementing the literature with such works will strengthen the argument that the blending variable optimization proposed in the article addresses a real and insufficiently described research gap.
2) It is essential to add 2-3 sentences that clearly and logically specify the research goal at the end of the literature review. Such additions will clearly demonstrate that the presented goal stems directly from the identified research gap and is not merely an intuitive proposition of the authors. This will make the argument structure more complete, and the motivation for the work clearer and more convincing to reviewers and readers. 3) It is essential to include at least two block diagrams: (1) a block diagram of the model architecture (Expert CR → Expert IS → Blending → Output) and (2) a simulation/validation loop diagram, showing the flow of information between the model modules and the signals representing the actuators' operation. Such illustrations will not only improve the readability of the work but will also clearly demonstrate to MDPI Actuators that the authors understand the system context in which the model actually supports the control of the actuators.
3) The paper lacks a complete and unambiguous description of the entire computational pipeline, including data acquisition, preprocessing, and the division into reference and test maneuvers. The authors also did not provide a data flow diagram or details regarding the sequence of calibration and validation steps, which makes it difficult to assess the consistency of the methodology. Supplementing this section with a clear diagram or detailed description will significantly improve the transparency and reproducibility of the research.

4) The description of the blending variable optimization process is too general and does not include key information regarding the algorithm used, parameters, and convergence criteria. Furthermore, data on the initialization method, sensitivity analysis, and potential numerical constraints are missing, making it difficult to assess the reliability of the results. Clarifying these elements would allow for a better assessment of the robustness of the optimization process and would increase its transparency.
5) The paper does not indicate the computational environment in which the analyses were conducted, nor the IT tools, numerical libraries, or software versions used. Information on the hardware configuration is also missing, making it impossible to assess the complexity of the calculations and potential limitations resulting from computational power. Providing this information will significantly improve the transparency of the research and ensure its full reproducibility.
6) The paper lacks formal cross-validation and tests on independent maneuvers, limiting the ability to assess the model's generalization beyond the dataset used for calibration. The authors also did not test the solution's robustness to measurement noise or variability in dynamic conditions, which is particularly important in the case of vehicle models. Such tests are essential to supplement the research. This will significantly enhance the reliability of the results and confirm the practical usefulness of the proposed approach.
7) The reviewed manuscript completely omitted analysis of measurement uncertainty and the impact of sensory errors, which constitutes a serious methodological gap in the context of vehicle dynamics modeling. Models of this type are inherently sensitive to input signal noise, sensor drift, and parameter variability, therefore, the lack of an assessment of their robustness undermines the reliability of the obtained results and limits their practical application. From a scientific perspective, it is necessary to present at least a basic sensitivity or uncertainty propagation analysis to demonstrate the stability of the models and confirm their adequacy for applications in real-world control systems. This should be presented in the Discussions section.
8) The reviewed manuscript lacks any quantitative reference to more advanced computational methods, such as modern neural models, PINN, or probabilistic methods, despite being discussed extensively in the literature review. The lack of even a basic comparative benchmark hinders the assessment of the relative quality of the proposed solution and makes it difficult to determine whether the benefits stem from the method itself or merely from the data selection. From a scientific perspective, a minimal comparison with a single contemporary method is necessary to reliably position the presented model in the current state of research. This should be presented in the Discussions section.

9) The final conclusions largely refer to the presented results, but not all of them are sufficiently closely linked to the experimental data obtained. Some statements, especially those concerning potential applications in control systems and the impact on actuator performance, are declarative in nature and do not directly derive from the tests conducted. A clear indication of the limitations of the research is also lacking, even though the results suggest certain areas of lower model stability. Supplementing the conclusions with these elements will allow for a more comprehensive empirical justification and will increase the coherence of the work with the conducted analyses.

Comments on the Quality of English Language

The English is correct but requires proofreading by a native speaker to achieve the natural fluency, precision, and stylistic consistency necessary for acceptance in an MDPI-recognized journal.

Author Response

Comment 1: It is worthwhile to add 2-3 literature sources from the LPV and SDRE areas for heavy vehicles, as this will allow for a stronger contextualization of the problem within the context of contemporary state-dependent control methods.

Response 1: We appreciate the suggestion to contextualize our work within the broader scope of Linear Parameter-Varying (LPV) and State-Dependent Riccati Equation (SDRE) methods. We have updated the Introduction to include key references from Vu et al., Tjønnås & Johansen, and Németh et al. These additions clarify that while these advanced frameworks provide robust theoretical foundations for adaptive control, they often lack a systematic methodology for the 'selection of the scheduling variable' under data-sparse conditions—a specific gap our study addresses through the proposed optimization-based blending strategy.

Reviewer Comment 2: It is essential to add 2-3 sentences that clearly and logically specify the research goal at the end of the literature review.

Response 2: We have revised the end of the Literature Review section to explicitly state the research goal. The added text clarifies that the primary objective is “to develop a methodologically robust blending architecture that enables parsimonious grey-box models to generalize across distinct physical regimes (steady-state vs. transient) without relying on complex black-box structures or noise-sensitive heuristic variables.”

Reviewer Comment 3:

It is essential to include at least two block diagrams: (1) a block diagram of the model architecture (Expert CR → Expert IS → Blending → Output) and (2) a simulation/validation loop diagram.

Response 3:

The figures presented below have also been incorporated into the relevant subsections of the Materials and Methods section.

Figure 1: Simulation/Validation Loop Diagram

Figure 2 : The block Diagram of Model Architecture

Reviewer Comment 4: The paper lacks a complete and unambiguous description of the entire computational pipeline, including data acquisition, preprocessing, and the division into reference and test maneuvers.

Response 4: vWe have expanded Section 2 to provide a rigorous description of the computational pipeline, ensuring reproducibility. Based on the related script, we specified the following steps:

Preprocessing: Raw signals were smoothed using a smoothing spline algorithm (smoothing parameter) to eliminate sensor quantization noise while preserving dynamic peaks.

Differentiation: Kinematic variables such as longitudinal acceleration () and steering rate () were derived analytically from the fitted spline objects to prevent noise amplification associated with numerical differentiation.

Maneuver Segmentation: Test vectors were automatically trimmed to start exactly when the steering input exceeds a threshold of 0.17 rad (10 degrees), ensuring temporal alignment across all validation scenarios.

Reviewer Comment 5: The description of the blending variable optimization process is too general and does not include key information regarding the algorithm used, parameters, initialization method, sensitivity analysis, and convergence criteria.

Response 5: To derive the robust blending weights demonstrated in Table 2, the following optimization logic was executed:

Input Space Definition: Three candidate state variables were normalized: Longitudinal Speed (), Steering Angle (δ_sw), or Steering Rate ()

Objective Function: A cost function was defined to minimize the deviation between the model's computed regime weight () and the ground-truth regime labels ( for steady-state, for transient) across the calibration dataset:

Solver Configuration: A constrained nonlinear optimization solver (fmincon in MATLAB) was employed with the Sequential Quadratic Programming (SQP) algorithm.

Convergence: The optimization converged to a solution where the coefficients for Speed and Steering Rate were negligible (<10^-3), while the Steering Angle coefficient was dominant. This data-driven process mathematically justifies the "Angle-Only" architecture, proving it is not an arbitrary choice but the result of minimizing regime identification error.

Reviewer Comment 6: The paper does not indicate the computational environment (IT tools, numerical libraries, software versions, hardware configuration).

Response 6: All simulations and optimizations were performed in MATLAB 2024a on a computer equipped with an AMD Ryzen 5 5600H 3.30 GHz processor, 16 GB RAM, and an NVIDIA GeForce GTX 1650 graphics card. The integration time step was fixed at (500 Hz), and the ODE45 solver was used with relative and absolute tolerances of and , respectively.

Reviewer Comment 7:

The paper lacks formal cross-validation and tests on independent maneuvers. The authors also did not test the solution's robustness to measurement noise or variability in dynamic conditions.

Response 7: We respectfully clarify that the manuscript does include rigorous cross-validation and robustness testing, specifically in Section 3.3 (Generalization) and Section 3.7 (Robustness).

Cross-Validation: We performed two distinct validation tests: 'Extrapolation' (Test 1) and 'Interpolation' (Test 2), where the model was trained on a specific subset of maneuvers and validated against strictly unseen maneuvers. The RMSE results in Table 4 quantify this generalization capability.

Noise Robustness: As detailed in Section 3.7 and supported by our Monte Carlo simulation script, we subjected the models to 50 iterations of Gaussian input noise. We have expanded the discussion of these results to explicitly address the reviewer's concern regarding sensor noise stability.

Reviewer Comment 8: Provide at least a basic sensitivity or uncertainty propagation analysis to demonstrate the stability of the models.

Response 8: To strictly address the stability and robustness concerns, we incorporated a comprehensive Monte Carlo Input Uncertainty Analysis in the new Section 3.7. By injecting Gaussian noise into the steering and speed inputs across 50 simulation runs, we mapped the output variance envelope of the Dynamic Model. The quantitative results (Tables 8 & 9) indicate that while the Dynamic Model exhibits a marginally higher output variance than the Static Model—an expected outcome of its adaptive gain-scheduling architecture—the standard deviation of the yaw rate error remains bounded ( rad/s). This induced variance is commensurate with the typical signal noise floor of automotive sensors and is negligible compared to the significant accuracy gains achieved. These findings confirm that the model maintains structural stability and is robust enough for real-world control applications where sensor noise is inevitable.

Reviewer Comment 9: The manuscript lacks any quantitative reference to more advanced computational methods (neural models, PINN, etc.). A minimal comparison is necessary.

Response 9: We explicitly considered comparing our approach against extensive black-box methods like Neural Networks (NN) or PINNs. However, such a comparison was deemed methodologically inconsistent with the core constraint of this study: 'Data Sparsity'.

Modern NN-based estimators typically require large, diverse datasets (thousands of data points covering the full slip envelope) to generalize without overfitting. In our industrial use-case, obtaining such limit-handling data is costly and hazardous. Our dataset consists of only four reference maneuvers. Training a neural network on such a sparse dataset would inevitably lead to severe overfitting, making it a methodologically unsound benchmark due to the lack of sufficient training data.

Therefore, we chose the 'Heuristic (Lateral Acceleration) Model' as the benchmark because it represents the actual industrial state-of-the-art alternative for low-order, tunable models used in ECUs today. Our results show that our 'Angle-Only' method outperforms this standard industrial heuristic, which is the relevant practical contribution. defined in our problem statement.

Reviewer Comment 10: The final conclusions need to be more closely linked to the experimental data, avoiding declarative statements about actuators not derived from tests. Limitations of the research must be clearly indicated.

Response 10: We have refined the Conclusions section to align strictly with the experimental evidence presented. Declarative statements regarding specific actuator hardware performance—which were not directly measured—have been removed. The conclusion now focuses on the quantified improvements in lateral acceleration tracking accuracy (17.7%) and the proven stability in limit-handling maneuvers (Figure 4), directly referencing the validation data. We have also added a dedicated paragraph explicitly stating the limitations of the current two-regime approach.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors This paper presents the methodological evolution of a foundational static vehicle model into an adaptive, robust, and physically plausible dynamic model. While the work is original and interesting, the authors need to address the following comments before it can be considered for publication at a reputed venue like 'Actuators: - My major concern is on its validation as well as real-world applicability. Include experimental results acquired using a physical system to prove the efficacy and usefulness of the proposed state-dependent blending methodology. - The model itself is a vague and needs to be more crisper, comprehensive, concrete and conclusive to convince its potential impact. - The discussion on models based on heuristics variables, mentioned on Line 72, could benefit from a work from literature such as 'Optimisation of fuzzy logic quadrotor attitude controller–particle swarm, cuckoo search and BAT algorithms'. - Elaborate clearly on the rationale of using parsimonious grey-box models. - Some mathematical symbols (see for example, Page 5) are not properly displayed. It could be an error during conversion into PDF. Please check and fix it. - Avoid using the first person in a research paper. There are so many 'We' in the manuscript. - Fuzzy control can be further improved through adaptation. Elaborate this further on Page 2 with reference to 'Hierarchical fuzzy-adaptive position control of an active mass damper for enhanced structural vibration suppression'. - Define all the abbreviations at their first occurrence and later on, simply use the abbreviation. - The paper needs thorough proofreading for typos and other linguistic improvements. Please remove the typos such as " alternatives,8 without".

Author Response

Reviewer Comment 1: Include experimental results acquired using a physical system to prove the efficacy and usefulness of the proposed state-dependent blending methodology.

Response 1: We thank the reviewer for emphasizing the importance of physical validation. Regarding the scope of validation and real-world applicability, we respectfully clarify that the primary objective of this study is to propose a methodological perspective on addressing the trade-off between 'data scarcity' and 'model generalization' in heavy vehicle dynamics.

In the current State-of-the-Art (SOA), developing high-fidelity parsimonious models often requires extensive datasets or scenario-specific re-calibration, which can be prohibitive due to the logistical and safety constraints of heavy vehicle testing. Our study aims to demonstrate that it is possible to develop regime-agnostic models—capable of adapting to distinct dynamic characteristics (steady-state vs. transient)—using only sparse experimental data.

By validating the proposed 'Angle-Only' blending architecture against experimental data collected on a closed test track (specifically Constant Radius and Increasing Steer maneuvers as detailed in Section 3), we illustrate that this methodology yields physically plausible and accurate simulations without the extensive tuning typical of traditional approaches. Therefore, the experimental validation in this work is intended to verify the predictive capability of this modeling framework as a robust foundation for future control applications, rather than to demonstrate a closed-loop implementation within the scope of this study. We have revised the Introduction to clearer articulate this focus on methodological robustness under data constraints.

Reviewer Comment 2: The model itself is vague and needs to be more crisper, comprehensive, concrete and conclusive.

Response 2: To address the vagueness, we have significantly restructured Section 2 (Material and Methods). We have added:

Explicit Equations: The exact mathematical formulation of the 5-parameter model and the cubic blending law (Eq. 7).

Block Diagrams: The following figures illustrating the model architecture and the simulation loop.

Figure 1: Simulation/Validation Loop Diagram

Figure 2 : The block Diagram of Model Architecture

Algorithm Details: A step-by-step description of the optimization logic used to derive the blending weights as follows. These additions make the model structure concrete and reproducible.

To derive the robust blending weights demonstrated in Table 2, the following optimization logic was executed:

Input Space Definition: Three candidate state variables were normalized: Longitudinal Speed (), Steering Angle (δ_sw), or Steering Rate ()

Solver Configuration: A constrained nonlinear optimization solver (fmincon in MATLAB) was employed with the Sequential Quadratic Programming (SQP) algorithm.

Reviewer Comment 3: The discussion on models based on heuristics variables could benefit from a work from literature such as 'Optimisation of fuzzy logic quadrotor attitude controller...'.

Response 3: We have incorporated the suggested study by Zatout et al. into the Introduction. This reference enriches our discussion on how optimization algorithms (such as Particle Swarm or BAT algorithms) are essential for tuning heuristic-based controllers, paralleling our own use of optimization to select the blending variable.

Reviewer Comment 4: Elaborate clearly on the rationale of using parsimonious grey-box models.

Response 4: We have expanded the Introduction to clearly articulate the rationale for parsimony. In heavy vehicle applications, data sparsity is a critical constraint; obtaining limit-handling data is costly and risky. High-fidelity models suffer from parameter unidentifiability when data is scarce, while black-box models (Neural Networks) lack physical interpretability and safety guarantees outside their training set. Our parsimonious grey-box approach is selected specifically to ensure identifiability from sparse data (only 4 maneuvers) and real-time computational efficiency (<1 ms execution time), which are prerequisites for embedded vehicle control systems.

Reviewer Comment 5: Some mathematical symbols are not properly displayed. Please check and fix it.

Response 5: We apologize for the display errors. We have thoroughly proofread the manuscript and re-entered all mathematical symbols using standard Equation Editors to ensure they render correctly in the final PDF.

Reviewer Comment 6: Avoid using the first person ("We") in a research paper.

Response 6: The manuscript has been meticulously revised to strictly adhere to the passive voice, eliminating first-person references as requested.

Reviewer Comment 7: Elaborate on Fuzzy control improvement through adaptation with reference to 'Hierarchical fuzzy-adaptive position control...'.

Response 7: We have added the reference by Saleem et al. to the Literature Review section. This citation supports our discussion on how hierarchical and adaptive mechanisms can significantly enhance the performance of baseline control structures in dynamic environments, providing a relevant precedent for our proposed adaptive blending architecture.

Reviewer Comment 8: Define all the abbreviations at their first occurrence and later on, simply use the abbreviation.

Response 8: We have reviewed the manuscript to ensure all abbreviations (e.g., ADAS, RMSE, IMU, CAN) are explicitly defined upon their first occurrence.

Reviewer Comment 9: The paper needs thorough proofreading for typos and other linguistic improvements (e.g., "alternatives,8").

Response 9: We thank the reviewer for the careful eye. The typo 'alternatives,8' and other linguistic errors have been corrected. The manuscript has undergone a thorough proofreading to improve flow and precision.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Comments and Suggestions for Authors

The authors have incorporated all suggested comments, leading to significant improvements in the revised version of the paper. It is recommended for acceptance.