In electric vehicles, regenerative braking converts kinetic energy into electrical energy during deceleration by switching the wheel motor into generator mode. BLDC motors are often used for their efficiency and dual motor/generator capability (Nian et al., 2014). Main system components include wheel motors, bidirectional power electronics (e.g., H-bridge inverters with MOSFET/IGBT switches), and energy storage units like lithium-ion batteries or supercapacitors. Braking triggers the motor to act as a generator, inducing back EMF that charges the battery via the bidirectional converter. Dual-motor all-wheel-drive EVs further enhance energy recovery compared to single-motor setups (Heydari et al., 2020).

Braking systems in electric vehicles are categorized into three: mechanical, regenerative, and hybrid. Mechanical braking relies on friction, turning kinetic energy into heat and causing wear, but provides crucial safety. Regenerative braking slows the vehicle by using the motor as a generator to recover energy, though its effectiveness is limited when the battery is fully charged or at low speeds. Hybrid systems integrate both methods, ensuring reliable deceleration across all conditions by compensating for the limitations of regenerative braking, especially during low speeds or full battery scenarios (Da and Bo, 2015). In emergency braking, the system defaults fully to friction braking for maximum safety. Thus, the best systems balance efficient energy recovery with uncompromised safety.

Regenerative braking systems must balance energy recovery with braking stability and safety. At low speeds, conversion efficiency drops significantly. Emergency stops require deceleration forces that exceed the capacity of regenerative systems. Frequent charge and discharge cycles also accelerate battery degradation. High current charging, especially at high temperatures or elevated state of charge (SOC), increases the risk of lithium plating and capacity loss. Over time, this cycling reduces battery storage ability and overall vehicle performance. Although regenerative braking can extend lithium-ion battery life by reducing deep discharges (Chidambaram et al., 2023), frequent high SOC charging especially in warm conditions can speed-up degradation. While regenerative braking mainly accelerates cycling aging, constant high SOC and elevated temperatures also drive calendar aging during storage (Keil and Jossen, 2017). To prevent damage, battery management systems must carefully regulate charging rates, SOC, and temperature to optimize both energy recovery and long-term battery health. Transition lag between mechanical and regenerative braking further complicates stability during rapid stops.

Polynomial Regression (PR) and Random Forest Regression (RFR) are used to optimize regenerative braking energy in electric vehicles. PR models braking force as a nonlinear function of inputs such as brake demand and battery SOC, using gradient descent for optimization and incorporating principal component analysis (PCA) to reduce dimensionality. RFR, which combines predictions from multiple decision trees trained on random data subsets, identifies brake demand as the important factor for accurate braking force prediction. RFR achieved better accuracy with a lower root mean square error (RMSE) of 5.16e-04 compared to PR's 0.042588 and handled nonlinear patterns more effectively. However, PR produced higher regenerative force in most driving conditions. For example, 16,000 W vs. 15,770 W in FTP cycles (Prasanth et al., 2023). Both models underperform relative to real-time optimization (RTO) methods but offer useful compromises: PR is faster and suitable for real-time use, whereas RFR manages complex nonlinearities better but with higher computational delay. A major limitation of PR is its depends on fixed polynomial terms, which can result in underfitting (Kim et al., 2021). Overall, both techniques improved energy recovery by approximately 59% compared to traditional fuzzy logic and neural network approaches (Prasanth et al., 2023).

Artificial neural networks (ANNs) predicts braking force by analyzing inputs like battery SOC, speed, and brake demand. ANNs minimizes prediction errors by continuous weight updation which enables the network to learn input-output relationships. Unlike traditional experimental or rule-based approaches, the ANN model offers faster, more flexible predictions and enables real-time control system development without physical trials (Rezk and Abuzied, 2023). The neural network (NN) architecture for braking force optimization is illustrated in Figure 3. The studies such as Velu and Chellammal (2023) introduce a novel Improved Wild Horse Optimization Algorithm (IWHO) to optimize ANN weights, achieving good accuracy. This approach maximizes energy recovery by 15% while ensuring vehicle stability through optimal brake force distribution. Models like Long Short-Term Memory (LSTM) and Nonlinear Autoregressive Exogenous (NARX) can predict how much energy can be recovered during braking (Ziadia et al., 2023). These models consider driver-specific regeneration limits and road topology into optimization, achieving 16%–39.6% energy recovery improvements across various driver cycles. It is expected that this adaptive approach can reduce computational load while outperforming traditional methods in downhill scenarios.

In Neural Inverse Optimal Control (NIOC), a special type of neural network called a Recurrent High-Order Neural Network (RHONN) is trained using an Extended Kalman Filter (EKF) (Ruz-Hernandez et al., 2022). This training helps the network understand how voltage and current behave in a buck-boost converter. This approach enables accurate tracking of time-varying references and optimizes energy recovery during braking without solving complex equations. It achieves high tracking accuracy and improves SOC retention by 20.41%. Neural network models like LSTM and Transformer help estimate energy use during braking by learning from how the vehicle is driven and the road conditions (Zhang et al., 2024). This helps in dynamically estimating regenerative braking efficiency, which traditional models often oversimplify with static coefficients while achieving nearly 15% lower errors.

However, ANN performance relies heavily on training data quality. It requires large datasets and face computational complexity and makes real-time deployment difficult. Additionally, the “black-box” nature of NN reduces interpretability compared to rule-based models (Zhang et al., 2024). Also for Optimization Algorithms like IWHO, computational complexity would make real-time implementation difficult. Although this method performs better than traditional fuzzy logic and many ANN approaches, it shows poor real-time optimization in dynamic energy recovery.

Deep Reinforcement Learning (DRL) is a method where an agent learns to make decisions by interacting with an environment. It uses deep neural networks to understand complex patterns and improve actions over time. The agent gets rewards for good actions and learns to avoid bad ones. Over many tries, it figures out the best way to reach a goal. DRL is being used in areas such as modern autonomous vehicles and related energy systems. For instance, Min et al. (2019) proposes a multi-level deceleration algorithm that blends driver modeling and optimization-based techniques to identify optimal deceleration trajectories. In another study (Kim et al., 2021), the regenerative braking process is modeled using parameterized polynomial deceleration and utilizes dynamic programming to maximize energy recapture during braking events. By integrating external factors such as traffic signals and road topography, the system dynamically adjusts deceleration commands, yielding at least a 16% increase in energy recovery and a 3%-10% reduction in trip time compared to human drivers, while maintaining a 3-meter safety margin in 95% of test cases. Enabling V2I (vehicle-to-infrastructure) communication can help in recovering more energy when braking near traffic lights (Zhang et al., 2020). Here the braking process is set up as a decision-making problem using RL, where the vehicle's distance and speed be the “state” and different ways of slowing down are the “actions.” The algorithm tries different actions, learns from the results, in order to maximize the rewards- balances energy recovery and smooth driving. With the available real-time traffic light data, two RL methods were tested in the studies, Q-Learning and Deep Q-Network(DQN). The Q-learning method recovered 45.08% more energy than simple uniform braking and DQN recovered 2.24% more energy and demonstrated smoother braking.

To resolve the inherent trade-offs between different system objectives (energy recovery, comfort, and safety), DRL algorithms commonly use a custom reward function to that quantifies these often conflicting objectives. One such representative reward function r_t at time step t is formulated as:

Here E_regen(t) denotes the energy recovered during braking, normalized by the maximum possible recoverable energy E_max. The term (adesired-at)2 penalizes deviations from the desired deceleration a_desired, promoting smooth and comfortable braking. The indicator function I(violation) imposes penalties for unsafe conditions, such as excessive wheel slip or insufficient safety margins. The weighting factors λ₁, λ₂, and λ₃ balance the trade-offs between maximizing energy efficiency, maintaining braking comfort, and ensuring safety.

This reward structure enables the DRL agent to dynamically adjust braking strategies, optimizing energy recapture while preserving vehicle stability and passenger comfort. It clearly expresses the key algorithmic foundation in many DRL-based regenerative braking systems.

Recent advancements demonstrate the diverse potential of DRL in this field. For example, the Deterministic Policy Gradient (DPG) approach has been applied to develop adaptive regenerative braking strategies for pure electric vehicles, leading to 57.69% faster convergence and nearly 2% lower energy consumption compared to traditional control, as well as extended battery life (Xu, 2025). Another framework (Ramesh et al., 2024) combines DRL with modular recurrent neural networks (RNNs) for tasks such as localization, path planning, and power management. The algorithm leverages real-time sensor and historical data, with modular RNNs reducing overall power consumption via time-series memory reuse, ultimately showing improved energy efficiency in highway scenarios. Further, the use of the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm enables dynamic allocation of braking torque across motors and hydraulic systems, delivering 35.18% energy recovery under various driving cycles (Peng and He, 2025). In more complex hybrid setups, Fuzzy Q-Learning (FQL) has been used to optimize hydraulic regenerative braking by dynamically adjusting force in real time, also introducing trauma memory to enhance safety in rare collision scenarios (Ning et al., 2023). These models achieved energy recovery up to 9.62% higher than standard fuzzy control and successfully maintained strict safety margins. Both Kim et al. (2021) and Ning et al. (2023) report 100% success in Euro NCAP AEB tests at 60 km/h, along with smooth, efficient deceleration profiles—a notable improvement over conventional rule-based approaches.

Fuzzy logic models work in a way that's similar to how humans make decisions. Membership functions in fuzzy logic is used to define how each input value is mapped to a degree of membership between 0 and 1. They help translate real-world inputs (like battery SOC, temperature, vehicle speed, or brake requirement) into fuzzy values like “low,” “medium,” or “high,” this process is also known as fuzzification. This allows the system to handle uncertainty and make decisions in a way that mimics human reasoning. The result is defuzzified to get a crisp output. Fuzzy logic is one of the most commonly used RBS techniques according to existing literature. Most of the papers use fuzzy logic as a method to improve the existing methods such as PID controllers, reinforcement learning or neural networks. The integration of fuzzy inference systems with conventional PID controllers for braking force optimization is shown schematically in Figure 4.

Advanced fuzzy logic techniques, such as Type-2 Fuzzy Logic (IT2FL) (Wahid et al., 2024), have been used to optimize regenerative braking in electric trikes, utilizing the Nie-Tan method as a type-reducer to streamline computation. This approach improved energy recovery by 3% (2.3 km range extension) over conventional fuzzy systems and delivered 35.84% total energy recovery (91 km increase) and 50 km/kWh efficiency (41.7% improvement) compared to non-regenerative control. Other works (Islam et al., 2023) integrate fuzzy logic with PID controllers, resulting in 8%–25% greater driving range and enhanced stability. In fuzzy models such as Maia et al. (2020) a technique called Batch Fuzzy Q-Learning (BFQL) is used to optimize the regenerative braking factor (RBF). Integrating this reinforcement learning technique eliminates the need for manual expert adjustments to automate the tuning of fuzzy rules. This approach improved energy recovery predictions, reducing errors by up to 22.47% compared to manual methods.

Gradient-based fuzzy logic controllers (Vodovozov et al., 2020) further adapt braking force to changing road conditions, facilitating smooth shifts between normal and anti-lock modes. Simulations report recovery of 22 kJ even under variable tire-road friction. Two-layer fuzzy-PID control strategies (Anh et al., 2024) separate front/rear braking force allocation (using ideal braking curves) from the real-time fuzzy distribution, yielding 13%–30% energy recovery gains. Takagi–Sugeno fuzzy sliding-mode controllers (TSFSMC) (Zhang et al., 2016) ensure above 91% efficiency without additional sensors by ensuring precise voltage regulation and fast responsiveness in dynamic operating conditions.

A two-layer fuzzy logic control (FLC) system (Yin et al., 2023a) employs 64 fuzzy rules and centroid defuzzification to dynamically determine braking ratios, with integrated PID controllers ensuring stable nonlinear dynamics and achieving up to 40% simulated energy savings. Hybrid approaches using PID controllers combined with Adaptive Neuro-Fuzzy Inference System (ANFIS) (Akhila and Ratnan, 2016) deliver rapid response and adaptability to changing conditions, resulting in a 50% increase in driving range and superior battery management vs. stand alone PID or FLC. Takagi–Sugeno (T-S) type fuzzy neural networks (FNN) (Li W. et al., 2024) can self-optimize fuzzy rules, enhancing flexibility, accuracy, and driving range by 19.2%. For real-time driving-mode switching, fuzzy logic-based decision systems (Ning et al., 2023) optimize transitions between single-motor and dual-motor modes, using an evaluation layer to minimize unnecessary switching and mechanical wear, while improving overall system stability.

A notable study on hybrid energy storage systems (HESS) combining supercapacitors and batteries (Li W. et al., 2024) demonstrates that a FLC can dynamically split braking energy, prioritizing the supercapacitor during high-energy events to reduce strain on battery. This control approach achieves a 29.1% reduction in battery current and a 46.84% decrease in heat generation, thus extending battery life. Separately, a fuzzy logic-based regenerative braking system using a Sugeno-type controller (Xu et al., 2011) introduces innovations like series braking (allowing independent control of the front and rear axles) and acceleration sensors that align braking distribution with ideal curves—effectively preventing wheel lock and enhancing stability. This system leads to a 25.7% increase in driving range and a 22.2% improvement in energy efficiency compared to vehicles without coordinated fuzzy braking control.

The major drawback of most of the fuzzy logic based RBS is the complexity of tuning membership functions, especially under variable and nonlinear driving conditions. Managing rapid SOC fluctuations while balancing real-time responsiveness with computational efficiency make real-world implementation challenging. In Şen et al. (2024) and Wahid et al. (2024) which has hybrid energy systems, managing the fast interaction between the super-capacitors and battery is also an important concern.

Genetic Algorithms (GAs) operate by evolving a population of candidate solutions through processes such as selection, crossover, and mutation, ultimately converging on optimal or near-optimal solutions. Unlike gradient-based methods, GAs are well-suited for complex, multi-modal problems where derivative information may not be available. In regenerative braking optimization, GAs evaluate and refine braking control settings across generations, prioritizing energy recovery and safety.

For example, a GA-based eco-driving approach for electric vehicles (Gautam et al., 2021) optimized acceleration profiles and cruising speeds, successfully minimizing energy consumption. This method effectively avoided local optima and managed computational complexity, outperforming a Stochastic Hill Climber (SHC) method by achieving greater energy savings. However, it faced challenges with the encoding of control variables as 14–32 bit chromosomes and with balancing search space efficiency. In another context, a GA-driven regenerative braking system for railways (Che et al., 2022) was developed to optimize power distribution through railway power networks and energy converters. In hardware-in-loop tests, this strategy realized 93.3% utilization of regenerative braking energy—significantly surpassing the SHC approach and reduced three-phase current imbalance by 98.6%, a critical improvement for railway systems.

Some studies propose hybrid optimization schemes combining Genetic Algorithms (GA) with fuzzy logic for enhanced regenerative braking. For instance, Bostanian et al. (2013) details a GA-tuned FLC strategy that optimizes the split of braking torque between mechanical and regenerative systems during deceleration, resulting in a 26% reduction in both energy costs and emissions compared to standard controllers. A key challenge was integrating 16 control settings into the GA, while ensuring vehicle performance constraints were still satisfied. For another approach, the hybrid neuro-fuzzy-genetic algorithm (NFGA) (Arunprasad et al., 2023), uses neural networks for system modeling, fuzzy logic for decision processes, and GA for optimal parameter tuning. This comprehensive framework achieved a 40% reduction in control deviation and substantially improved trajectory tracking accuracy, demonstrating the value of combining soft computing and evolutionary techniques in RBS control.

Swarm intelligence based approaches for regenerative braking uses biologically inspired collective problem-solving to optimize energy recovery. In Particle Swarm Optimization (PSO), each “particle” evaluates candidate solutions and moves according to both its own historical best and the swarm's collective best, iteratively converging on optimal braking force allocation. Ant Colony Optimization (ACO) uses artificial “ants” to explore potential strategies, with solution quality guiding future searches via virtual pheromone trails. A recent study (Chai et al., 2022) applied PSO to tune a PI controller for regenerative braking in electric vehicles using super-capacitors. This optimization produced a 6% increase in travel distance, a 14.57% rise in average speed, a 3.69% improvement in maximum speed, and lower speed tracking errors. However, the increased energy recovery demanded careful management of super-capacitor SOC, which dropped by 1.58% due to higher usage. The PSO-selected controller parameters (K_p and K_i) also enhanced BLDC motor performance during braking, maintaining voltage stability (with a constant 540V initial supercapacitor voltage) throughout deceleration events.

The paper (Pradhan et al., 2023) proposes a PSO-optimized PID controller for brushless DC (BLDC) motor speed control in electric vehicles. This method aimed at improving regenerative braking efficiency such that it automatically tunes the PID gains (K_p, K_i, K_d) using PSO, resulting in faster settling time, reduced speed tracking errors, and a 3% reduction in SOC loss during braking compared to conventional PID tuning. Figure 5 illustrates the overall architecture of a PSO optimized PID controller, highlighting how the PSO algorithm dynamically adjusts PID parameters in response to real-time driving conditions. Such AI-based adaptive control yields improved energy recovery, quicker response to disturbances, and enhanced robustness over traditional control strategies, as confirmed by recent simulation results.

Another approach employs PSO in combination with fuzzy logic to optimize regenerative braking force distribution in automobiles (Tuan et al., 2020). This method achieved a 10.49%–24.44% reduction in fuel consumption by maximizing both the duration and efficiency of regenerative braking, particularly surpassing conventional fuzzy logic strategies in low-speed scenarios. Further advancements are seen in the use of swarm intelligence-based hybrid strategies that combine PSO and ACO to optimize braking torque distribution in hybrid electric vehicles (Zhang et al., 2022, 2019). Specifically, Zhang et al. (2019) reports a 16.04% increase in energy recovery compared to rule-based methods. Key findings include a 63.51% regenerative energy recovery rate in urban driving cycles (Zhang et al., 2022), outperforming both PSO-only (62.26%) and rule-based (54.24%) approaches. The principal advantage of these hybrid PSO-ACO strategies lies in their ability to avoid local optima, ensuring stable slip ratios (around 0.2) during emergency braking, while simultaneously balancing considerations related to battery aging and safety.

Studies undoubtedly shows that AI and ML based models outperform conventional rule-based methods in optimizing battery recharge. For example, a fuzzy logic-based regenerative braking strategy achieved a 22.2% increase in energy efficiency and a 25.7% improvement in driving range over conventional rule-based systems (Xu et al., 2011). Reinforcement learning (Q-learning) and deep Q-networks dynamically optimize braking actions further. This capability further improves energy recovery over static approaches (Yin et al., 2024).

AI based systems, especially the models that use neural networks and fuzzy logic is observed to ensure smoother deceleration by intelligently distributing braking force. An ANN based comfort regenerative braking system reduced acceleration jerk from 0.35 g/s to 0.05 g/s, providing a much smoother ride than conventional methods (Hwang et al., 2023). Fuzzy logic controllers also minimize abrupt braking transitions, enhancing passenger comfort (Yin et al., 2023a).

AI approaches adapt to terrain, vehicle load, and driver behavior dynamically. For instance, a BLSTM neural network model manipulates regenerative braking in the best possible way for individual driving styles, improving energy recovery by up to 16.3% for aggressive drivers and 9.8% for moderate drivers over traditional strategies (Chengqun et al., 2023). LSTM and NARX models predict braking needs over long horizons, adapting to changing road gradients and traffic conditions (Ziadia et al., 2023).

AI-based systems enhance stability and safety by maintaining optimal braking force distribution, which helps prevent wheel lock and ensures vehicle control even on slippery roads (Vodovozov et al., 2021). AI and ML algorithms help preserve battery health by regulating charging rates during regenerative braking, which reduces stress on the battery and extends its lifespan (Chidambaram et al., 2023). Although AI systems deliver superior performance, they demand greater computational resources than simpler rule-based logic, which is easier to implement but less effective in handling complex, real-world scenarios (Prasanth et al., 2023).

Therefore, considering factors such as energy efficiency, braking comfort, adaptability, stability, and battery health, -based techniques are preferable. However, the computational demands associated with -based techniques present a significant challenge.

The hybrid AI framework proposed for electric vehicle (EV) regenerative braking systems is an integrated control architecture that combines the strengths of several AI and ML techniques specifically, fuzzy logic, neural networks, reinforcement learning, genetic algorithms, and swarm intelligence. The intent is to leverage the complementary properties of these techniques to suggest a robust, real-time, and adaptive optimization of both energy recovery and vehicle safety in highly variable road and traffic environments.

As demonstrated in both the manuscript's comparative table and quantitative summaries, individual methods (e.g., fuzzy logic, neural networks, reinforcement learning) each have strengths (such as adaptability, computational speed, or optimality in non-linear systems) but also inherent limitations such as poor real-time response, need for extensive training data, or lack of interpretability. Hybridization allows each component method to operate in domains where it excels, while other methods compensate for its weaknesses. Figure 6 represents the proposed Hybrid AI Framework for regenerative braking in electric vehicles. The framework integrates multiple AI and ML techniques including neural networks for perception and prediction, fuzzy logic for handling uncertainty, reinforcement learning for adaptive control, and genetic/swarm intelligence for parameter optimization to dynamically and robustly optimize braking force distribution. The layered architecture processes real-time vehicle and environmental data to maximize energy recovery while ensuring vehicle stability and safety under varying road conditions.

The proposed Hybrid AI Framework is structured into three key layers and a feedback loop:

By integrating data-driven perception, rule-based reasoning, and learning-based policy refinement, the framework adapts to different drivers, road conditions, and vehicle states more effectively than any standalone method. The system has demonstrated significant improvements in energy recovery, with hybrid fuzzy-neural, GA-fuzzy, and PSO-ACO optimized methods achieving up to 63 percent recovery, especially in mixed and urban driving conditions. The fuzzy logic and reinforcement learning modules work together to ensure that emergency handling and vehicle stability are never compromised in the pursuit of efficiency. The architecture is designed to be easily extended, allowing for the integration of future AI advancements such as explainable AI for greater transparency and edge computing layers to reduce latency.

Previous studies, including those using hybrid fuzzy-neuro-genetic and PSO-ACO optimizers, have demonstrated the effectiveness of modular hybrid controllers; however, real-world implementation and benchmarking in electric vehicles are still limited. Future work should prioritize developing lightweight, real-time AI models, conducting field trials with automotive partners, establishing unified benchmarks comparing rule-based and single-AI systems, and integrating transparent explainability modules to ensure safety compliance.

AI/ML strategies in RBS face significant obstacles regarding hardware limitations. Deep learning (DL) and reinforcement learning (RL) models require considerable computational resources for training and real-time inference. For example, Q-learning-based braking optimization needs continuous data processing to adapt to dynamic driving conditions. While edge computing and AI chips are emerging as solutions, integrating these with existing EV architectures is technically complex and costly, hindering widespread adoption.

To be more effective, training AI/ML models requires large and high-quality datasets. The variability of driving behavior, road conditions, and state of charge (SOC) during a drive makes it difficult to collect comprehensive data to optimize regenerative braking. LSTM-based approaches for energy prediction, for instance, face accuracy issues due to insufficient real-world data. Noisy or incomplete sensor data can further degrade model performance, leading to inefficient braking allocation. Unlike traditional rule-based systems, AI techniques must balance regenerative and friction braking while dynamically adapting to drive conditions.

Optimizing for energy recovery sometimes compromises braking safety. Fuzzy logic controllers adjust braking torque based on road adhesion, but unpredictable conditions like sudden transitions between dry and icy roads can destabilize the vehicle. During emergency stops, the system must prioritize stopping power over energy harvesting.

Frequent regenerative braking cycles subject EV batteries to rapid charging and discharging, accelerating degradation, internal resistance, and heat generation. This undermines the benefits of regenerative braking by shortening battery lifespan. Reinforcement learning methods have been explored to optimize charge–discharge patterns, but real-time implementation remains difficult. Hybrid energy storage systems can reduce stress on a single battery, though they add complexity to AI-driven control strategies.

Most EVs use embedded systems such as anti-lock braking (ABS) and electronic stability control (ESC) operating on fixed protocols. Adding AI/ML-based RBS requires smooth, low-latency integration with these existing setups. For example, neural network controllers must work seamlessly with hydraulic actuators without performance lags. From the literature reviewed in this paper, one of the biggest challenges is the lack of standardized studies. Because researchers use different testing cycles and protocols, the reported energy recovery results vary, making it hard to compare them fairly. Additionally, these systems are vulnerable to cyberattacks, such as data poisoning and sensor spoofing, potentially leading to misallocated braking force and accidents. Moreover, existing studies vary widely in simulation environments, datasets, and evaluation metrics. Establishing standardized test protocols and datasets is essential to facilitate reproducible research and industry adoption.

The integration of AI and ML into regenerative braking systems promises significant advancements in EV efficiency and performance. However, challenges like computational limits, data quality, safety trade-offs, and regulatory gaps must be addressed to unlock this potential. Moving forward, future research should target these challenges through collaborative efforts among automakers, technology developers, and regulators. Such cooperation will be crucial for establishing standardized benchmarks, enabling large-scale deployment, and ultimately paving the way for smarter, safer, and more sustainable electric mobility.