Pith. sign in

REVIEW 3 major objections 4 minor 42 references

Self driving algorithm for an active four wheel drive racecar

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper reports that an end-to-end reinforcement-learning policy trained with PPO can control an active four-wheel-drive racecar by directly commanding steering and per-wheel torques, implicitly learning torque-vectoring and stability…

desk verdict A credible RL torque-vectoring demo whose headline comparison to physics-based controllers is unsupported by the experiments. read the letter →

arxiv 2506.06077 v1 pith:J4ZYHUBG submitted 2025-06-06 cs.RO

Gergely Bari , Laszlo Palkovics This is my paper
classification cs.RO
keywords deepreinforcementlearningproximalpolicyoptimizationtorquevectoringactivefour-wheeldriveautonomousracingvehicledynamicscontrolTORCSend-to-end
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper asks whether one neural network can replace the layered, physics-based control stack of an electric racecar with active four-wheel drive. It trains a PPO agent in the TORCS simulator that outputs only a steering angle and four independent wheel torques, and shows the agent learns to distribute torque corner-by-corner the way a torque-vectoring controller would. This matters because it suggests reinforcement learning can discover coordinated vehicle-dynamics control, including understeer mitigation, ABS, and traction control, without separate controllers or explicit models. The authors report lap times competitive with a passive-4WD baseline and measurably better combined grip usage during braking-and-turn-in phases.

What carries the argument

The load-bearing element is the end-to-end policy: a multilayer perceptron with three shared hidden layers (300, 600, 600 neurons) that maps a state vector of velocities, accelerations, yaw rate, wheel speeds, track geometry, and lidar into a five-dimensional action vector, one steering command and one torque per wheel, normalized to [-1, 1]. The agent is trained with PPO on a progress-based reward measuring distance along the track centerline, and the resulting torque patterns are the mechanism through which the paper claims the agent implicitly learns torque vectoring and stability control.

What would settle it

A transfer test would settle the claim: run the trained policy on the same vehicle dynamics in a high-fidelity simulator with a detailed combined-slip tire model (or on a physical A4WD test vehicle) and measure whether the inside-front-wheel spin and the mid-corner maximum-steering behavior still produce competitive lap times, or whether they cause overheating, wear, or grip loss that makes the policy slower than a physics-based torque-vectoring controller.

Watch

Extended reading notes

Core claim

The central claim is that an end-to-end deep reinforcement learning policy, trained with Proximal Policy Optimization, can learn time-optimal control of an active four-wheel-drive racecar by directly outputting a steering angle and independent torque commands for all four wheels. Analyzing the trained policy corner by corner, the paper reports the agent dynamically shifts torque between inside and outside wheels to counteract the vehicle's understeer, applies braking predominantly to the rear wheels in some corners, and reproduces yaw-stability behaviors analogous to ABS, ASR, and torque vectoring without those functions being encoded. The paper presents this as evidence that RL can serve as a unified alternative to classical vehicle-dynamics-control architectures for grip-limited driving.

Load-bearing premise

The result stands on the assumption that TORCS's simplified tire model is faithful enough that behaviors learned there, such as intentionally spinning the inside front wheel or using maximum steering in a hairpin, are genuine control strategies rather than artifacts that a real tire would punish.

Editorial extensions

If this is right

  • A single policy can simultaneously solve path following, stabilization, and torque allocation for an A4WD racecar, removing the need for separate vehicle-dynamics-control modules.
  • The learned torque patterns adapt across corner types on the same track, showing the policy changes torque distribution with corner phase rather than memorizing a single strategy.
  • The agent implicitly reproduces ABS, ASR, and yaw-moment control from reward alone, implying that stability functions could emerge from end-to-end learning rather than explicit design.
  • Because electric motor characteristics affect results, matching powertrain characteristics is essential when comparing RL control strategies against physics-based baselines.
  • A 20 Hz control frequency is enough to learn competitive racing behavior, and the paper notes that a higher control frequency could yield further gains from finer torque modulation.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The reported 25% higher combined g-forces during braking and turn-in suggests the learned policy uses independent wheel torque authority to approach the tire friction ellipse more aggressively than the passive 4WD baseline; a natural next step, which the paper does not include, is a direct comparison against an explicit model-predictive torque-vectoring controller on the same active vehicle model.
  • If the learned strategies are confirmed on higher-fidelity tire models, the same end-to-end formulation could be extended to other actuators such as active suspension or rear steering, and to sim-to-real pipelines with domain randomization.
  • The paper's own caveat about the simplified tire model implies a testable hypothesis: in a simulator with tire wear and thermal effects, the inside-wheel spin strategy would become suboptimal, and the agent would need a penalty or a richer tire model to rediscover a less abusive strategy.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper trains a PPO-based deep RL agent to control an active four-wheel-drive (A4WD) racecar in the TORCS simulator, with an action space consisting of steering angle and four independent wheel torques, and an observation space of vehicle dynamic states. The agent is trained with a progress-based reward plus termination and action penalties, and its learned behavior is analyzed corner-by-corner, including steering traces, wheel torque patterns, wheel speeds, and GG-diagrams. The authors report that the agent learns to mitigate understeer through differential torque application, implicitly mimics ABS/ASR-like behaviors, and achieves lap times competitive with a passively controlled 4WD vehicle. The paper claims that the learned policy mimics and potentially surpasses traditional physics-based A4WD controllers, though no such controller is implemented or benchmarked.

Significance. If the empirical findings are fully supported, the paper would be a useful demonstration that a single end-to-end RL policy can coordinate steering and per-wheel torque commands for limit-handling driving, potentially reducing the need for layered vehicle dynamics controllers. The paper is transparent about many training details, provides extensive visual and quantitative analysis of learned behaviors, and includes appendices with hyperparameters and turn-by-turn figures. The main scientific value is the evidence that DRL can discover non-trivial torque-vectoring strategies that mitigate understeer and manage yaw dynamics in a simulated racing scenario. However, the significance is substantially tempered by the absence of a physics-based A4WD baseline and by the acknowledged exploitation of TORCS's simplified tire model, which leaves the central comparative claim unsupported and raises questions about transferability.

major comments (3)
  1. [Abstract; Section 4; Figure 4] The claim that the learned policy 'mimics and, in aspects of grip utilization, potentially surpasses traditional physics-based A4WD controllers' is not supported by the reported experiments, because no physics-based A4WD controller is implemented or benchmarked. The only comparison is against a passive 4WD vehicle controlled by steering and pedals (Figure 4), and the lap-time difference is confounded by the powertrain difference between the electric motors of the active car and the engine of the passive car, as the authors themselves acknowledge in Section 4. To substantiate the headline claim, the paper needs either a classical torque-vectoring A4WD controller evaluated on the same active vehicle model, or an ablation in which the torque action space is constrained (e.g., equal torques across wheels) while keeping the powertrain identical.
  2. [Section 4; Figure 4] The evaluation relies on single representative laps for each agent, with no reporting of variance across training seeds, evaluation episodes, or repeated runs. The approximately 25% higher combined g-forces and the lap-time advantage are based on one lap each, so it is unclear whether these differences are statistically meaningful rather than the result of selecting a favorable lap. The paper should report mean and standard deviation over multiple evaluation episodes (e.g., 10–20 deterministic runs) and, ideally, across multiple training seeds, to support quantitative claims of improvement.
  3. [Section 4, Turn 2 analysis] The paper states that the inside-front-wheel spin during corner exit 'incurs no penalty in TORCS's simplified model and might be exploited by the agent if it aids rotation or acceleration.' This admission directly undermines the interpretation of the learned torque strategy as a transferable A4WD control law. Since the central claim involves implicit learning of real-world-relevant control logic (ABS, ASR, torque vectoring), the manuscript must either (a) provide evidence that the learning is robust to penalizing such simulator-specific exploitation, e.g., by adding a slip penalty or using a more realistic tire model, or (b) substantially temper the claims of generalizable A4WD control and reframe the contribution as a simulation-specific demonstration.
minor comments (4)
  1. [Section 3.3, Eq. (2)] The action penalty in Eq. (2) is unclear: with the given parameters psc=15 and pbnd=1.2, the expression (|a|/psc - pbnd + 1)^2 yields a small but always positive penalty even for actions within the normalized [-1,1] range, which contradicts the stated purpose of discouraging actions outside the feasible range. Please clarify whether actions are clipped before this penalty and what the intended shape of the penalty is.
  2. [Section 4, Turn 2 text] There are several typos and grammatical errors that impede readability, including 'TurIn' (likely 'turn-in'), 'cars thends to US' (likely 'cars tend to understeer'), and a stray ']' in 'based on their characteristics]'. These should be corrected.
  3. [Table B2] Table B2 lists 'ASR_ONFalse' twice, which appears to be a formatting error. Additionally, the table would benefit from a note clarifying which settings correspond to the passive and active vehicle models, respectively.
  4. [Appendix Tables B1–B3] The appendix tables are introduced as 'T able B1' etc. in the captions, which is a rendering inconsistency. Also, the observation vector in Table B3 includes LiDaR signals and 'Episode dist.', but the main text describes the observation as consisting of vehicle dynamic states; please clarify whether LiDaR and episode distance are actually used as inputs.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the RL training result is measured independently of its reward proxy, and the only self-citation is a non-load-bearing contrast with prior work.

full rationale

The paper's central result is an empirical reinforcement-learning outcome: a PPO policy maps vehicle states to steering and four wheel-torque actions, and lap times are measured directly by the TORCS environment rather than derived from the reward function. The progress reward is a training proxy, but no fitted parameter is reused as a prediction, and no equation in the paper defines the output in terms of the input or vice versa. The one self-citation (the authors' precursor work, [37]) is used only to state what is different about the present approach and is not load-bearing for any conclusion. Claims that the learned torque patterns 'mimic' classical A4WD behaviors are interpretive analyses of recorded signals, not circular reductions; the absence of a physics-based A4WD baseline is an experimental-comparison limitation, not a circularity. The paper's own caveat that TORCS's simplified tire model may allow the agent to exploit unrealistic slip penalties is an honest limitation statement, and it strengthens rather than weakens the finding that the learning itself is not circular. No self-definition, fitted-input-as-prediction, or self-citation-chain pattern exists in the derivation chain.

Assumptions & free parameters 5 free parameters · 3 assumptions · 0 invented entities

The central claim is empirical, so the ledger is dominated by the simulator model, the modified powertrain model, and reward-shaping choices. No fit-to-data parameters are re-used as predictions, so circularity burden is low.

free parameters (5)
  • Action penalty scaling psc = 15
    Hand-chosen reward-shaping parameter (Table B1) controlling the penalty on actions outside [-1,1].
  • Action penalty boundary pbnd = 1.2
    Hand-chosen threshold in the action penalty (Table B1); tunes how strongly out-of-range actions are discouraged.
  • Termination rewards = +100 / -10
    Hand-chosen reward magnitudes for lap completion and failure modes (Table B1), which influence training behavior.
  • Learning rate schedule = [1:2.5, 0:0.5] * 10^-4
    PPO hyperparameter (Table B2) chosen to stabilize training, not derived from measurements.
  • Observation scaling factors = e.g., /300, /80, /pi, /100
    Hand-chosen normalizations (Table B3) to bring state measurements into comparable ranges for the neural network.
assumptions (3)
  • domain assumption TORCS simulator's vehicle and tire dynamics are a valid testbed for evaluating grip-limit driving control.
    The entire empirical evaluation runs in TORCS with the built-in Car7 model and physics (cite [21]); no validation against real vehicle data is provided.
  • domain assumption The modified active 4WD car model, with per-wheel electric motor torque characteristics, adequately represents an A4WD electric powertrain.
    Motor curves are introduced in Figure 2 with only a verbal description ('simple electric motor characteristics') and are not fully specified numerically.
  • ad hoc to paper Progress along the track centerline is a faithful surrogate for minimizing lap time.
    The reward (Eq. 1) uses centerline progress rather than true lap-time minimization; the paper relies on this proxy to provide a dense learning signal.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Self driving algorithm for an active four wheel drive racecar." pith.science (2026). https://pith.science/paper/J4ZYHUBG

@misc{pith2026250606077,
  author       = {Pith},
  title        = {Pith review of: Self driving algorithm for an active four wheel drive racecar},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/J4ZYHUBG}},
  note         = {Machine review of arXiv:2506.06077}
}
read the original abstract

Controlling autonomous vehicles at their handling limits is a significant challenge, particularly for electric vehicles with active four wheel drive (A4WD) systems offering independent wheel torque control. While traditional Vehicle Dynamics Control (VDC) methods use complex physics-based models, this study explores Deep Reinforcement Learning (DRL) to develop a unified, high-performance controller. We employ the Proximal Policy Optimization (PPO) algorithm to train an agent for optimal lap times in a simulated racecar (TORCS) at the tire grip limit. Critically, the agent learns an end-to-end policy that directly maps vehicle states, like velocities, accelerations, and yaw rate, to a steering angle command and independent torque commands for each of the four wheels. This formulation bypasses conventional pedal inputs and explicit torque vectoring algorithms, allowing the agent to implicitly learn the A4WD control logic needed for maximizing performance and stability. Simulation results demonstrate the RL agent learns sophisticated strategies, dynamically optimizing wheel torque distribution corner-by-corner to enhance handling and mitigate the vehicle's inherent understeer. The learned behaviors mimic and, in aspects of grip utilization, potentially surpass traditional physics-based A4WD controllers while achieving competitive lap times. This research underscores DRL's potential to create adaptive control systems for complex vehicle dynamics, suggesting RL is a potent alternative for advancing autonomous driving in demanding, grip-limited scenarios for racing and road safety.

Figures

Figures reproduced from arXiv: 2506.06077 by the authors.

Figure 1
Figure 1. Scheme of the Markov Decision Process, formalizing the Agent-Environment interaction in RL problems [13] [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Characteristics used to model wheel motors for the active 4WD car. 3.2. Observation and Action Space The Observation provided to the agent consists of a vector of vehicle dynamic states, detailed in Table B3. The Action space for the active 4WD agent comprises five con￾tinuous values normalized to the range [-1, 1]: one for the Steering Wheel Angle (SWA) (conventional front steering, +1 max left, -1 max right) and o… view at source ↗
Figure 3
Figure 3. shows lap times as a function of the training steps. The blue curve represents the lap times achieved during training episodes where actions were sampled stochas￾tically, while the red curve shows the lap times during evaluation episodes (test laps) using deterministic actions (the mean of the policy network’s output distribution). An interesting observation is that while the agent learns to navigate the track relat… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Comparison of representative laps between the agent controlling the passive 4WD vehicle (via steering/pedals) and the agent controlling the active A4WD vehicle (via steering/individual torques). Plots show (top to bottom): steering wheel angle, throttle/brake pedal (pa…
Figure 5
Figure 5. Figure 5: Detailed driving behavior of the active A4WD agent in Turn 1. Includes steering comparison with the passive 4WD agent for reference. line. Given this context, the agent’s strategy might be optimal within TORCS. The yaw rotation instability during braking is controlled …
Figure 6
Figure 6. Figure 6: Detailed driving behavior of the active A4WD agent in Turn 2 (slow hairpin). position the car for the subsequent short straight by not running wide on exit aligns with lap time optimization under these specific simulated constraints [PITH_FULL_IMAGE:figures/full_fig_p…
Figure 7
Figure 7. Figure 7: Detailed driving behavior of the active A4WD agent in Turn 3. tires, allowing them to generate higher longitudinal forces. The agent exploits this by demanding more braking force from the outer wheels on entry and more acceleration from them on exit. However, this opti…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

42 extracted references · 36 canonical work pages

  1. [1]

    Coordinating Tire Forces to Avoid Obstacles Using Nonlin- ear Model Predictive Control

    Brown M, Gerdes JC. Coordinating Tire Forces to Avoid Obstacles Using Nonlin- ear Model Predictive Control. IEEE Transactions on Intelligent Vehicles. 2020 Mar; 5(1):21–31. Conference Name: IEEE Transactions on Intelligent Vehicles; Available from: 14 https://ieeexplore.ieee.org/document/8910387/?arnumber=8910387

    work page arXiv 2020

  2. [2]

    Control Informed Design of the IAC Autonomous Racecar for Operation at the Dynamic Envelope

    Zhu Q, Schmid M, Prucka R, et al. Control Informed Design of the IAC Autonomous Racecar for Operation at the Dynamic Envelope ; 2024. ArXiv:2407.17737 [eess]; Available from: http://arxiv.org/abs/2407.17737

    work page Pith review arXiv 2024

  3. [3]

    Autonomous Vehicles on the Edge: A Survey on Au- tonomous Vehicle Racing

    Betz J, Zheng H, Liniger A, et al. Autonomous Vehicles on the Edge: A Survey on Au- tonomous Vehicle Racing. IEEE Open Journal of Intelligent Transportation Systems. 2022;3:458–488. ArXiv:2202.07008 [cs]; Available from: http://arxiv.org/abs/2202.07008

    arXiv 2022

  4. [4]

    Motion control in an electric vehicle with four independently driven in-wheel motors

    Sakai S, Sado H, Hori Y. Motion control in an electric vehicle with four independently driven in-wheel motors. IEEE/ASME Transactions on Mechatronics. 1999 Mar;4(1):9–16. Available from: http://ieeexplore.ieee.org/document/752079/

    work page 1999

  5. [5]

    UOT electric march II

    Yoichi Hori, Hori Y. Future vehicle driven by electricity and Control-research on four- wheel-motored "UOT electric march II". IEEE Transactions on Industrial Electronics. 2004 Oct;51(5):954–962. MAG ID: 2019708972

    work page 2004

  6. [6]

    Lateral-Stability-Oriented Path-Tracking Control Design for Four- Wheel Independent Drive Autonomous Vehicles with Tire Dynamic Characteristics under Extreme Conditions

    Yu Z, Zhao R, Yuan T. Lateral-Stability-Oriented Path-Tracking Control Design for Four- Wheel Independent Drive Autonomous Vehicles with Tire Dynamic Characteristics under Extreme Conditions. World Electric Vehicle Journal. 2024 Oct;15(10):465. Number: 10 Publisher: Multidisciplinary Digital Publishing Institute; Available from: https://www. mdpi.com/2032...

    work page 2024

  7. [7]

    Optimum traction force distribution for stability improvement of 4WD EV in critical driving condition

    He P, Hori Y. Optimum traction force distribution for stability improvement of 4WD EV in critical driving condition. In: 9th IEEE International Workshop on Advanced Motion Control, 2006.; Mar.; 2006. p. 596–601. ISSN: 1943-6580; Available from: https: //ieeexplore.ieee.org/abstract/document/1631727

    work page arXiv 2006

  8. [8]

    Research on Robust Control for longitudinal Impact of 4 Wheel-Drive Hybrid Electric Vehicle

    Xiong L, Yu Z. Research on Robust Control for longitudinal Impact of 4 Wheel-Drive Hybrid Electric Vehicle. Procedia Engineering. 2011 Jan;15:293–297. Available from: https://www.sciencedirect.com/science/article/pii/S187770581101558X

    work page 2011

Show all 42 references
  1. [9]

    Time Optimal Control of Four-in-Wheel-Motors Driven Electric Cars

    Max G, Lantos B. Time Optimal Control of Four-in-Wheel-Motors Driven Electric Cars. Periodica Polytechnica Electrical Engineering and Computer Science. 2014;58(4):149–159. Number: 4; Available from: https://pp.bme.hu/eecs/article/view/7806

    2014

  2. [10]

    Model predictive control allocation for stability improvement of four-wheel drive electric vehicles in critical driving condition

    Zhao H, Ren B, Chen H, et al. Model predictive control allocation for stability improvement of four-wheel drive electric vehicles in critical driving condition. IET Control Theory & Applications. 2015;9(18):2688–2696. _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1049/ie...

    2015

  3. [11]

    Learning-Based Model Predictive Control for Autonomous Racing

    Kabzan J, Hewing L, Liniger A, et al. Learning-Based Model Predictive Control for Autonomous Racing. IEEE Robotics and Automation Letters. 2019 Oct;4(4):3363–3370. Available from: https://ieeexplore.ieee.org/document/8754713/

    2019

  4. [12]

    GoodfellowI,BengioY,CourvilleA.Deeplearning.MITpressCambridge;2016.Available from: https://www.deeplearningbook.org/

    2016

  5. [13]

    Reinforcement learning: An introduction

    Sutton RS, Barto AG. Reinforcement learning: An introduction. MIT press; 2018

    2018

  6. [14]

    Dota 2 with Large Scale Deep Reinforcement Learning ; 2019

    OpenAI, :, Berner C, et al. Dota 2 with Large Scale Deep Reinforcement Learning ; 2019. Version Number: 1; Available from: https://arxiv.org/abs/1912.06680

    2019 arXiv

  7. [15]

    Grandmaster level in StarCraft II using multi-agent reinforcement learning

    Vinyals O, Babuschkin I, Czarnecki WM, et al. Grandmaster level in StarCraft II using multi-agent reinforcement learning. Nature. 2019 Nov;575(7782):350–354. Number: 7782 Publisher: Nature Publishing Group; Available from: https://www.nature.com/articles/ s41586-019-1724-z

    2019

  8. [16]

    Human-level control through deep reinforcement learning

    Mnih V, Kavukcuoglu K, Silver D, et al. Human-level control through deep reinforcement learning. nature. 2015;518(7540):529–533. Publisher: Nature Publishing Group

    2015

  9. [17]

    Solving Rubik’s Cube with a Robot Hand

    OpenAI, Akkaya I, Andrychowicz M, et al. Solving Rubik’s Cube with a Robot Hand

  10. [18]

    Emergence of Locomotion Behaviours in Rich Environ- ments ; 2017

    Heess N, TB D, Sriram S, et al. Emergence of Locomotion Behaviours in Rich Environ- ments ; 2017. ArXiv:1707.02286 [cs]; Available from: http://arxiv.org/abs/1707.02286

    2017 arXiv

  11. [19]

    Soft Actor-Critic: Off-Policy Maximum Entropy DeepReinforcementLearningwithaStochasticActor.In:Proceedingsofthe35thInterna- tional Conference on Machine Learning; Jul

    Haarnoja T, Zhou A, Abbeel P, et al. Soft Actor-Critic: Off-Policy Maximum Entropy DeepReinforcementLearningwithaStochasticActor.In:Proceedingsofthe35thInterna- tional Conference on Machine Learning; Jul. PMLR; 2018. p. 1861–1870. ISSN: 2640-3498; 15 Available from: https://pr...

    2018

  12. [20]

    Proximal Policy Optimization Algorithms ; 2017

    Schulman J, Wolski F, Dhariwal P, et al. Proximal Policy Optimization Algorithms ; 2017. ArXiv:1707.06347 [cs]; Available from: http://arxiv.org/abs/1707.06347

    2017 arXiv

  13. [21]

    Available from: https://sourceforge

    TORCS - The Open Racing Car Simulator ; 2020. Available from: https://sourceforge. net/projects/torcs/

    2020

  14. [22]

    Minimum lap time trajectory optimisation of per- formance vehicles with four-wheel drive and active aerodynamic control

    de Buck P, Martins JRRA. Minimum lap time trajectory optimisation of per- formance vehicles with four-wheel drive and active aerodynamic control. Ve- hicle System Dynamics. 2023 Aug;61(8):2103–2119. Publisher: Taylor & Francis _eprint: https://doi.org/10.1080/00423114.2022.210...

    2023

  15. [23]

    Contextual Tuning of Model Predictive Control for Autonomous Racing

    Fröhlich LP, Küttel C, Arcari E, et al. Contextual Tuning of Model Predictive Control for Autonomous Racing. In: 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS); Oct.; 2022. p. 10555–10562. ArXiv:2110.02710 [cs]; Available from: http://arxiv.org...

    2022 arXiv

  16. [24]

    Two-timescale Mechanism-and-Data-Driven Control for Aggressive Driving of Autonomous Cars ; 2023

    Lu Y, Yang B, Mo Y. Two-timescale Mechanism-and-Data-Driven Control for Aggressive Driving of Autonomous Cars ; 2023. ArXiv:2109.05170 [cs]; Available from: http://arxiv. org/abs/2109.05170

    2023 arXiv

  17. [25]

    Sim-to-real: Quadruped Robot Control with Deep Reinforcement Learning and Parallel Training

    Jiang H, Chen T, Cao J, et al. Sim-to-real: Quadruped Robot Control with Deep Reinforcement Learning and Parallel Training. In: 2022 IEEE International Confer- ence on Robotics and Biomimetics (ROBIO); 2022. p. 489–494. Available from: https: //ieeexplore.ieee.org/document/10011921

    2022

  18. [26]

    DeepGait: Planning and Control of Quadrupedal Gaits Using Deep Reinforcement Learning

    Tsounis V, Alge M, Lee J, et al. DeepGait: Planning and Control of Quadrupedal Gaits Using Deep Reinforcement Learning. IEEE Robotics and Automation Letters. 2020 Apr; 5(2):3699–3706. Conference Name: IEEE Robotics and Automation Letters; Available from: https://ieeexplore.iee...

    2020

  19. [27]

    Practicability analy- sis of online deep reinforcement learning towards energy management strat- egy of 4WD-BEVs driven by dual-motor in-wheel motors

    Zhiyan Feng, Qingang Zhang, Yiming Zhang, et al. Practicability analy- sis of online deep reinforcement learning towards energy management strat- egy of 4WD-BEVs driven by dual-motor in-wheel motors. Energy. 2023;S2ID: 80fc325cb4162bac42946dede3b3eab429a02fff

    2023

  20. [28]

    Physics-Guided Reinforcement Learning System for Realistic Vehicle Active Suspension Control ; 2024

    Nhu AN, Le NA, Li S, et al. Physics-Guided Reinforcement Learning System for Realistic Vehicle Active Suspension Control ; 2024. ArXiv:2408.08425; Available from: http://arxiv. org/abs/2408.08425

    2024 arXiv

  21. [29]

    Deep reinforcement learning based direct torque control strategy for distributed drive electric vehicles considering active safety and energy saving performance

    Wei H, Zhang N, Liang J, et al. Deep reinforcement learning based direct torque control strategy for distributed drive electric vehicles considering active safety and energy saving performance. Energy. 2022 Jan;238:121725. Available from: https://www.sciencedirect. com/science...

    2022

  22. [30]

    Deep Reinforcement Learning-Based Torque Vector- ing Control Considering Economy and Safety

    Deng H, Zhao Y, Lin F, et al. Deep Reinforcement Learning-Based Torque Vector- ing Control Considering Economy and Safety. Machines. 2023 Apr;11(4):459. Num- ber: 4 Publisher: Multidisciplinary Digital Publishing Institute; Available from: https: //www.mdpi.com/2075-1702/11/4/459

    2023

  23. [31]

    Path Tracking Control for Four-Wheel Independent Steering and Driving Vehicles Based on Improved Deep Reinforcement Learning

    Hua X, Zhang T, Cheng X, et al. Path Tracking Control for Four-Wheel Independent Steering and Driving Vehicles Based on Improved Deep Reinforcement Learning. Tech- nologies.2024Nov;12(11):218.Number:11Publisher:MultidisciplinaryDigitalPublishing Institute; Available from: http...

  24. [32]

    Champion-level drone racing us- ing deep reinforcement learning

    Kaufmann E, Bauersfeld L, Loquercio A, et al. Champion-level drone racing us- ing deep reinforcement learning. Nature. 2023 Aug;620(7976):982–987. Number: 7976 Publisher: Nature Publishing Group; Available from: https://www.nature.com/articles/ s41586-023-06419-4

    2023

  25. [33]

    On learning racing policies with re- inforcement learning ; 2025

    Czechmanowski G, Węgrzynowski J, Kicki P, et al. On learning racing policies with re- inforcement learning ; 2025. ArXiv:2504.02420 [cs]; Available from: http://arxiv.org/abs/ 2504.02420

    2025 arXiv

  26. [34]

    Towards Optimal Head-to-head Autonomous Racing with Curriculum Reinforcement Learning ; 2023

    Kalaria D, Lin Q, Dolan JM. Towards Optimal Head-to-head Autonomous Racing with Curriculum Reinforcement Learning ; 2023. ArXiv:2308.13491 [cs]; Available from: http: //arxiv.org/abs/2308.13491. 16

    2023 arXiv

  27. [35]

    ArXiv:2412.03803 [cs]; Available from: http://arxiv.org/abs/2412.03803

    SubositsJ,LeeJ,ManuelS,etal.TowardsanAutonomousTestDriver:High-Performance Driver Modeling via Reinforcement Learning ; 2024. ArXiv:2412.03803 [cs]; Available from: http://arxiv.org/abs/2412.03803

    2024 arXiv

  28. [36]

    Version Number: 1; Available from: https://arxiv.org/abs/2410.00358

    BockmanJ, HoweM,OrensteinA,etal.AARK:AnOpenToolkitforAutonomousRacing Research ; 2024. Version Number: 1; Available from: https://arxiv.org/abs/2410.00358

    2024 arXiv

  29. [37]

    Vision based driving agent for race car simulation environments

    Bári G, Palkovics L. Vision based driving agent for race car simulation environments

  30. [38]

    Stable-Baselines3: Reliable Reinforcement Learning Im- plementations

    Raffin A, Hill A, Gleave A, et al. Stable-Baselines3: Reliable Reinforcement Learning Im- plementations. Journal of Machine Learning Research. 2021;22(268):1–8. Available from: http://jmlr.org/papers/v22/20-1364.html

    2021

  31. [39]

    Super-Human Performance in Gran Turismo Sport Using Deep Reinforcement Learning

    Fuchs F, Song Y, Kaufmann E, et al. Super-Human Performance in Gran Turismo Sport Using Deep Reinforcement Learning. arXiv:200807971 [cs]. 2020 Aug;ArXiv: 2008.07971; Available from: http://arxiv.org/abs/2008.07971

    2020 arXiv

  32. [40]

    Outracing champion Gran Turismo drivers with deep reinforcement learning

    Wurman PR, Barrett S, Kawamoto K, et al. Outracing champion Gran Turismo drivers with deep reinforcement learning. Nature. 2022 Feb;602(7896):223–228. Available from: https://www.nature.com/articles/s41586-021-04357-7. 17 Appendix A. List of Acronyms Appendix B. Training param...

    2022

  33. [2019]

    ArXiv:1910.07113 [cs, stat]; Available from: http://arxiv.org/abs/1910.07113

    1910 arXiv

  34. [2025]

    ArXiv:2504.10266 [cs]; Available from: http://arxiv.org/abs/2504.10266

    arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.