REBot: Reflexive Evasion Robot for Instantaneous Dynamic Obstacle Avoidance

Ce Hao; Chunzheng Wang; Fan Shi; Jin Song Dong; Kuankuan Sima; Zihao Xu

REVIEW 4 major objections 5 minor 41 references

REBot: Reflexive Evasion Robot for Instantaneous Dynamic Obstacle Avoidance

T0 review · 4 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read A learned reflex, not trajectory replanning, lets a quadruped avoid obstacles with under 1.5 seconds of warning.

desk verdict Useful control-system contribution with an honest but load-bearing perception assumption; deserves review, not unconditional acceptance. read the letter →

arxiv 2508.06229 v1 pith:UVH3EXCD submitted 2025-08-08 cs.RO

Zihao Xu , Ce Hao , Chunzheng Wang , Kuankuan Sima , Fan Shi , Jin Song Dong This is my paper · ORCID

classification cs.RO

keywords quadrupedalrobotsdynamicobstacleavoidancereflexiveevasionreinforcementlearningfinite-statemachinerecoverycontrolsim-to-realUnitreeGo2

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

The paper proposes that quadrupedal robots can handle obstacles that arrive too fast for path replanning by acting reflexively, and that the reflex can be learned. REBot is a finite-state machine with three stages: normal operation, a learned avoidance policy that dodges the incoming obstacle, and a learned recovery policy that regains balance afterwards. The authors train both policies with a standard policy-gradient method in a GPU simulator, using a two-stage curriculum from suddenly appearing static obstacles to moving ones, plus regularization and adaptive rewards. They report that this split—evade first, recover second—raises avoidance success over navigation-based baselines when reaction time is below 1.5 seconds, and that the same policies transfer to a real quadruped. If the system is right, instantaneous dynamic obstacle avoidance does not require replanning speed; it requires a fast evasive skill and a separate stabilization skill.

What carries the argument

The load-bearing object is the three-stage finite-state machine of the REBot system. The Avoidance stage is driven by a policy-gradient-trained actor network that outputs joint target angles; the Recovery stage is a second trained network activated by thresholds on body orientation, joint velocity, and base height. The FSM's work is to decouple the reflex from balance restoration: the avoidance policy can prioritize speed at the expense of stability, while the recovery policy restores normal function. A two-stage curriculum—static obstacle appearing at random position, then moving obstacle on a directed trajectory—and reward terms for trot symmetry, energy, contact smoothness, diversity, and

What would settle it

Run the same REBot policies on a quadruped using only onboard sensing, removing the motion-capture ground-truth injection, and measure avoidance success at reaction times between 0.5 and 1.5 s; if the success rate falls to the level of the navigation baselines, the claimed reflexive evasion was an artifact of perfect state information.

Watch

Extended reading notes

Core claim

The paper's central claim is that reflexive evasion should be a distinct control layer for legged robots: when reaction time is under 1.5 s, the robot should not try to recompute a safe path, it should immediately execute a learned evasive gesture. REBot implements this as a finite-state machine. On detecting an obstacle moving toward the robot, it enters the Avoidance stage; a neural-network policy trained by reinforcement learning maps proprioceptive state plus obstacle position and velocity to joint targets, producing behaviors like jumping backward for frontal threats and crouching for overhead ones. When the dodge leaves the body tilted, joint velocities extreme, or base height low, REB

Load-bearing premise

The robot receives exact ground-truth position and velocity of itself and the obstacles at every control step from a motion-capture system, so the central evasion result assumes such perception is available; the paper leaves precise obstacle perception to future work.

Editorial extensions

If this is right

At reaction times between 0.5 and 1.5 s, REBot reports an avoidance success rate of about 0.65, versus 0.11 and 0.09 for the ABS and reactive-RL baselines, so a reflex layer can outperform replanning exactly where replanning is hardest.
Removing the recovery stage substantially reduces post-avoidance stabilization in the reflex regime, so the evasive dodge and the balance recovery should be designed together, not as one black-box policy.
The learned policy produces direction-dependent behaviors—jumping away for frontal and lateral threats, crouching for overhead ones—so one reactive controller can cover a family of evasive maneuvers rather than a single predefined motion primitive.
In the longer reaction-time regime (1.5–4.0 s), REBot still beats the baselines while using less maximum joint power and less base displacement, indicating the reflex architecture does not sacrifice planned, efficient avoidance.
The reported real-robot success rates (ASR 56%, RSR 53%) show the trained policies transfer beyond simulation, making the framework a candidate for safe legged robots in human environments.

Reading between the lines

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

Extension: replacing the motion-capture ground truth with onboard perception is the natural next stress test; REBot's control layer is agnostic to where obstacle state comes from, so the perception module, not the reflex policy, is the remaining bottleneck.
Extension: the observed asymmetry—backward jumps are easier for the Go2's leg morphology—suggests morphology-aware action priors or asymmetric rewards could balance evasion capability across directions for other quadruped designs.
Extension: the reflex-versus-navigation split implies a hybrid controller that uses reaction time or threat level to switch between REBot's evasion policy and a replanning layer, which would extend the finite-state machine to continuously varying threat levels.
Extension: a falsifiable prediction of the design is that failures in a mixed-speed obstacle course should concentrate in perception or actuation, not in the decision layer, because the FSM explicitly covers the short-reaction-time regime.

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.

Desk Editor's Note

read the letter

The real contribution here is the FSM control structure: an avoidance policy and a recovery policy trained with a two-stage curriculum and adaptive rewards, targeting sub-1.5s reaction times. That specific integration is new for quadrupedal reflexive evasion, and the simulation ablations give it credible support. Removing the recovery stage drops success rates by about 20% in the reflex regime; removing the adaptive reward collapses gait diversity; the curriculum helps. The authors also report real hardware trials, which is more than many RL locomotion papers bother with, and they state their limitations plainly. That honesty is worth crediting.

The soft spots are real and roughly as the reader framed them. The baselines are weak: ABS is not built for dynamic obstacles, and RRL is a UAV policy with simplified rigid-body dynamics, so beating them mostly shows the problem is hard, not that REBot is strong. The real-world ASR of 56% and RSR of 53% sit uneasily with the word "robust" in the abstract. There are no error bars, no trial counts, no code or data, and several key thresholds and reward coefficients are undisclosed.

The biggest issue is the perception assumption. Appendix III says the robot receives ground-truth position and velocity for itself and the obstacle at every control step from OptiTrack, and the conclusion concedes that precise obstacle perception is left as an assumption. Since the avoidance policy and the FSM trigger consume p_o and v_o directly, the entire reflexive-evasion claim is conditional on having perfect state. No experiment varies observation noise or latency, and the 56% hardware ASR already hints that the policy is fragile to reality. The stress-test note lands: the paper's central promise is a robot that reflexively avoids in real time, but what is actually validated is a control policy that works given oracle perception. That is a meaningful distinction.

Still, the paper is not incoherent and does not overclaim relative to its own evidence. It is a solid empirical study of a control architecture, with honest limitations and a reproducible-in-principle simulation setup.

My recommendation: send it to peer review. A good referee will push for code or data, for more hardware trials, for stronger baselines, and for at least one experiment with noisy or delayed perception. If those are addressed, the paper would be a useful reference for legged-robot safety. As it stands, I would not put it in a reading group as a must-read, and I would not cite it in my own work until the perception gap is closed.

Referee Report

4 major / 5 minor

Summary. The paper introduces REBot, a finite-state-machine (FSM) control framework for quadrupedal robots that combines a reinforcement-learned avoidance policy and a recovery policy to handle dynamic obstacles under short reaction times (<1.5 s). The avoidance policy is trained in Isaac Gym with a two-stage curriculum (static then moving obstacles) and includes avoidance, regularization, and adaptive reward components; the recovery policy stabilizes the robot after evasive maneuvers. The authors evaluate REBot in simulation against ABS and RRL baselines, report avoidance success rate (ASR), recovery success rate (RSR), maximum joint power (MJP), and avoidance moving distance (AMD), and conduct ablations removing the recovery stage, first curriculum, and adaptive reward. A real-robot demonstration on a Unitree Go2 uses OptiTrack motion capture to feed ground-truth robot and obstacle states directly to the controller; real-world ASR is reported as 56% and RSR as 53%.

Significance. If the claims held fully, REBot would be a useful step toward reactive, reflex-like obstacle avoidance for legged robots, complementing navigation-based replanning methods. The paper's strengths include a clearly formulated problem, a structured FSM integrating two learned policies, ablation studies substantiating the contributions of the recovery stage, curriculum learning, and adaptive reward, and honest acknowledgment of limitations in the conclusion. The empirical evidence is, however, weakened by the reliance on ground-truth state injection from an external motion-capture system in the real-robot experiments, by the absence of statistical error bars or trial counts, and by the underspecification of reward coefficients and FSM thresholds. As the real-world success rates are far below simulation, the 'robust' and 'real-time' system-level claims are not yet established. These issues are load-bearing but appear fixable with additional experiments and reporting, hence the recommendation for major revision.

major comments (4)

[Appendix III, Sec. VI, Conclusion] The real-robot demonstration injects ground-truth position and velocity of both the robot and the obstacle from OptiTrack at every control step. The conclusion states that precise obstacle perception is left as an assumption. Because the FSM trigger in Sec. IV-A and the avoidance policy consume p_o and v_o, the central real-time DOA result is conditional on an oracle perception layer. No experiment varies observation noise or latency, and Table V randomizes obstacle initial conditions rather than measurement noise. Please either add robustness experiments with noisy or delayed obstacle observations or explicitly scope the contribution as control-only in the abstract and introduction.
[Tables I and II, Sec. VI] All simulation metrics are reported as single numbers without confidence intervals, numbers of trials, or random seeds. The ablation differences are small (e.g., ASR 0.65 vs. 0.59 for 'w/o adp.' in the 0.5–1.5 s row) and may be within run-to-run variance. The real-robot ASR (56%) and RSR (53%) also lack trial counts or per-scenario breakdowns. Please provide statistics over multiple seeds and trials, and specify how success is judged in hardware experiments.
[Sec. IV-B and Appendix I] Reward coefficients for r_avoidance, r_regularization, and r_adaptive are never specified, nor are the hyperparameters λ and η in the threat reward r_threat. The FSM instability thresholds θ_th, qdot_th, and h_th in Sec. IV-A are also absent. This prevents reproduction of the training pipeline and makes the claim of 'carefully designed' rewards unverifiable. Please list all reward weights and threshold values, or release the training code.
[Sec. V-B, Table I] The baselines ABS and RRL are not adapted to the task: ABS targets static obstacles and RRL targets UAV dynamics. A comparison against a legged reactive planner or an RL policy trained on the same observation space without the FSM/recovery stage would better support the claim that the proposed structure, rather than the training setup, yields the improvements. The ablations in Table II partly address this, but the abstract's 'notable improvements' rests on a weaker baseline comparison.

minor comments (5)

[Sec. IV-A] The transition condition for switching to the avoidance stage is garbled by missing symbols: '�v� , p� � p� � > �' should be the dot product of obstacle velocity and relative position. Please fix the equation rendering.
[Sec. IV-B] Several equations contain placeholder symbols, e.g., rdistance = � �� d(p_o, OBB) − r_o� and rcollision = ��c �� c ��. Please replace with standard mathematical notation.
[Sec. V-C] The metric 'MJD' appears where the text should refer to 'MJP' (maximum joint power) or 'AMD' (avoidance moving distance); the boundary description 'MJD below 300 W' indicates a typo. Also ensure consistent use of MJP/AMD throughout Figures and text.
[Sec. VI] There is a missing space before 'Under the real-world test conditions' in the last paragraph. Additionally, the sentence 'the REBot system achieves an ASR of 56% and an RSR of 53%' would benefit from specifying the total number of trials and the definition of recovery success on hardware.
[Fig. 5 caption] The red dashed region labeled 'reflexive evasion' should be defined in the caption; the reader must infer the reaction-time intervals from Table I.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: REBot is an empirical RL system whose claims rest on independent simulation and hardware metrics, not on a fitted parameter or self-referential derivation.

full rationale

The paper is an empirical reinforcement-learning systems paper, not a first-principles derivation, and its load-bearing claims are evaluated by external success/failure metrics: avoidance success rate (ASR), recovery stability rate (RSR), maximum joint power (MJP), avoidance moving distance (AMD), and gait diversity index (GDI), measured on rollouts in Isaac Gym, sim-to-sim transfer to MuJoCo, and a real Unitree Go2. The avoidance policy is trained with hand-designed rewards that include distance-to-obstacle and collision terms, but success is not defined as the reward value; it is a binary collision-free outcome under the signed-distance-function criterion. The reported hardware ASR of 56% and RSR of 53% are independent of any fitted parameter. There is no fitted-input-called-prediction step: no parameter is fit to a subset of the evaluation data and then used to predict that same data. There is also no uniqueness theorem or load-bearing mathematical claim imported from the authors' prior work; the only self-citations are peripheral motivation and RL references ([1], [17]) and are not load-bearing. Appendix III's explicit statement that the real robot 'directly receives ground truth position and velocity information for both itself and the obstacles at each control step' via OptiTrack, together with the conclusion's admission that 'we leave precise obstacle position perception as an assumption,' is a genuine limitation and a correctness/generalization risk, but it is not circularity: the control policy is empirically measured under that stated assumption, not defined by it, and the paper does not present perceived-state performance as a derived result. Therefore the derivation chain is self-contained in the sense relevant to circularity: no claimed result reduces, by construction or by self-citation, to its own inputs.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the reliability of simulation training, ground-truth perception, and hand-tuned reward coefficients. The paper provides no code, so the exact coefficients and thresholds are unknown. No new physical entities are introduced.

free parameters (4)

Reward coefficients for ravoidance, rregularization, radaptive
Weights combining distance, collision, walk, energy, contact, diversity, threat, and direction rewards are not given numerical values; the paper reports only functional forms.
Threat reward hyperparameters lambda, eta
Defined in the r_threat equation but values are never stated.
FSM instability thresholds theta_th, qdot_th, h_th
Criteria in Sec. IV-A name thresholds for orientation, joint velocity, and base height but do not report chosen values.
Region boundary thresholds in Sec. V-C = ASR 30%, MJD 300 W
Used to partition behavior space into failure, reflex, and navigation regions; thresholds appear selected post hoc to fit the observed data.

assumptions (4)

domain assumption Isaac Gym simulation dynamics and contact models are faithful enough to transfer to MuJoCo and real hardware
Entire training pipeline in Sec. IV-D and Appendix II depends on simulator fidelity; real-world performance drops to 56% ASR confirm this is an assumption.
domain assumption Ground-truth obstacle and robot state is available at each control step
Appendix III uses OptiTrack motion capture; conclusion states precise obstacle position perception is left as an assumption. The policy receives perfect states, not on-board perception.
domain assumption PPO with stated hyperparameters and curricula converges to a policy that generalizes across obstacle directions, speeds, and reaction times
The paper presents empirical results but provides no convergence or generalization guarantee; ablations support the design but do not prove generality.
ad hoc to paper Reward shaping terms (regularization and adaptive rewards) induce natural, stable, and energy-efficient behavior
The reward forms in Sec. IV-B are designed by the authors to elicit specific behaviors; there is no external benchmark validating that these rewards are necessary or optimal.

how reviews work

0 comments

Cite this review

Pith. "Pith review of REBot: Reflexive Evasion Robot for Instantaneous Dynamic Obstacle Avoidance." pith.science (2026). https://pith.science/paper/UVH3EXCD

@misc{pith2026250806229,
  author       = {Pith},
  title        = {Pith review of: REBot: Reflexive Evasion Robot for Instantaneous Dynamic Obstacle Avoidance},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UVH3EXCD}},
  note         = {Machine review of arXiv:2508.06229}
}

read the original abstract

Dynamic obstacle avoidance (DOA) is critical for quadrupedal robots operating in environments with moving obstacles or humans. Existing approaches typically rely on navigation-based trajectory replanning, which assumes sufficient reaction time and leading to fails when obstacles approach rapidly. In such scenarios, quadrupedal robots require reflexive evasion capabilities to perform instantaneous, low-latency maneuvers. This paper introduces Reflexive Evasion Robot (REBot), a control framework that enables quadrupedal robots to achieve real-time reflexive obstacle avoidance. REBot integrates an avoidance policy and a recovery policy within a finite-state machine. With carefully designed learning curricula and by incorporating regularization and adaptive rewards, REBot achieves robust evasion and rapid stabilization in instantaneous DOA tasks. We validate REBot through extensive simulations and real-world experiments, demonstrating notable improvements in avoidance success rates, energy efficiency, and robustness to fast-moving obstacles. Videos and appendix are available on https://rebot-2025.github.io/.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 40 canonical work pages

[1]

Re- thinking robustness assessment: Adversarial attacks on learning-based quadrupedal locomotion controllers,

F. Shi, C. Zhang, T. Miki, J. Lee, M. Hutter, and S. Coros, “Re- thinking robustness assessment: Adversarial attacks on learning-based quadrupedal locomotion controllers,” �� , 2024

work page 2024
[2]

Mobile robot collision avoidance based on deep reinforcement learning with motion constraints,

Y . Tao, M. Li, X. Cao, and P. Lu, “Mobile robot collision avoidance based on deep reinforcement learning with motion constraints,” �� , 2024

work page 2024
[3]

Spark: A modular benchmark for humanoid robot safety,

Y . Sun, R. Chen, K. S. Yun, Y . Fang, S. Jung, F. Li, B. Li, W. Zhao, and C. Liu, “Spark: A modular benchmark for humanoid robot safety,”

work page
[4]

Dynamic obstacle avoidance for quadrotors with event cameras,

D. Falanga, K. Kleber, and D. Scaramuzza, “Dynamic obstacle avoidance for quadrotors with event cameras,” �� , vol. 5, no. 40, p. eaaz9712, 2020

work page 2020
[5]

Learning a state representation and navigation in cluttered and dynamic envi- ronments,

D. Hoeller, L. Wellhausen, F. Farshidian, and M. Hutter, “Learning a state representation and navigation in cluttered and dynamic envi- ronments,” �� , vol. 6, no. 3, pp. 5081–5088, 2021

work page 2021
[6]

Robust autonomous navigation of a small-scale quadruped robot in real-world environments,

T. Dudzik, M. Chignoli, G. Bledt, B. Lim, A. Miller, D. Kim, and S. Kim, “Robust autonomous navigation of a small-scale quadruped robot in real-world environments,” in �� . IEEE, 2020, pp. 3664–3671

work page 2020
[7]

Learning vision-guided quadrupedal locomotion end-to-end with cross-modal transformers,

R. Yang, M. Zhang, N. Hansen, H. Xu, and X. Wang, “Learning vision-guided quadrupedal locomotion end-to-end with cross-modal transformers,” �� , 2021. �� Fig. 8: REBot system real-robot demonstrations on Unitree Go2 Robot (See video). (a) the robot is poked from different directions using a stick; (b) a ball is launche...

work page 2021
[8]

Creating a dynamic quadrupedal robotic goalkeeper with reinforcement learning,

X. Huang, Z. Li, Y . Xiang, Y . Ni, Y . Chi, Y . Li, L. Yang, X. B. Peng, and K. Sreenath, “Creating a dynamic quadrupedal robotic goalkeeper with reinforcement learning,” in �� . IEEE, 2023, pp. 2715–2722

work page 2023

Show all 41 references

[9]

Agile but safe: Learning collision-free high-speed legged locomotion,

T. He, C. Zhang, W. Xiao, G. He, C. Liu, and G. Shi, “Agile but safe: Learning collision-free high-speed legged locomotion,” �� , 2024

2024
[10]

Fapp: Fast and adaptive perception and planning for uavs in dynamic cluttered environments,

M. Lu, X. Fan, H. Chen, and P. Lu, “Fapp: Fast and adaptive perception and planning for uavs in dynamic cluttered environments,” �� , 2024

2024
[11]

Robust and versatile bipedal jumping control through reinforcement learning,

Z. Li, X. B. Peng, P. Abbeel, S. Levine, G. Berseth, and K. Sreenath, “Robust and versatile bipedal jumping control through reinforcement learning,” �� , 2023

2023
[12]

Future spinal reflex is embedded in primary motor cortex output,

T. Umeda, O. Yokoyama, M. Suzuki, M. Kaneshige, T. Isa, and Y . Nishimura, “Future spinal reflex is embedded in primary motor cortex output,” �� , vol. 10, no. 51, p. eadq4194, 2024

2024
[13]

Deployment of whole-body locomotion and manipulation algorithm based on nmpc onto unitree go2quadruped robot,

M. Liu, J. Xiao, and Z. Li, “Deployment of whole-body locomotion and manipulation algorithm based on nmpc onto unitree go2quadruped robot,” in �� . IEEE, 2024, pp. 1–6

2024
[14]

Egocentric visual locomotion in a quadruped robot,

F. Xiao, T. Chen, and Y . Li, “Egocentric visual locomotion in a quadruped robot,” in �� , 2024, pp. 172–177

2024
[15]

Prox- imal policy optimization algorithms,

J. Schulman, F. Wolski, P. Dhariwal, A. Radford, and O. Klimov, “Prox- imal policy optimization algorithms,” �� , 2017

2017
[16]

Reinforcement learning for quadrupedal locomotion: Current advancements and future perspectives,

M. Gurram, P. K. Uttam, and S. S. Ohol, “Reinforcement learning for quadrupedal locomotion: Current advancements and future perspectives,” in �� . IEEE, 2025, pp. 28–38

2025
[17]

Dexterous legged locomotion in confined 3d spaces with reinforcement learning,

Z. Xu, A. H. Raj, X. Xiao, and P. Stone, “Dexterous legged locomotion in confined 3d spaces with reinforcement learning,” in �� . IEEE, 2024, pp. 11 474–11 480

2024
[18]

Isaac gym: High performance gpu-based physics simulation for robot learning,

V . Makoviychuk, L. Wawrzyniak, Y . Guo, M. Lu, K. Storey, M. Macklin, D. Hoeller, N. Rudin, A. Allshire, A. Handa �� , “Isaac gym: High performance gpu-based physics simulation for robot learning,” �� , 2021

2021
[19]

Cerberus in the darpa subterranean challenge,

M. Tranzatto, T. Miki, M. Dharmadhikari, L. Bernreiter, M. Kulkarni, F. Mascarich, O. Andersson, S. Khattak, M. Hutter, R. Siegwart �� , “Cerberus in the darpa subterranean challenge,” �� , vol. 7, no. 66, p. eabp9742, 2022

2022
[20]

Path planning of forest fire-fighting robots based on deep learning

S.-j. Sun, S.-h. Jiang, S.-h. Cui, Y . Kang, and Y .-t. Chen, “Path planning of forest fire-fighting robots based on deep learning.” 2020

2020
[21]

Trot gait stability control of small quadruped robot based on mpc and zmp methods,

X. Meng, W. Liu, L. Tang, Z. Lu, H. Lin, and J. Fang, “Trot gait stability control of small quadruped robot based on mpc and zmp methods,” ��, vol. 11, no. 1, p. 252, 2023

2023
[22]

Adaptive gait planning for quadruped robot based on center of inertia over rough terrain,

J. Chen, K. Xu, and X. Ding, “Adaptive gait planning for quadruped robot based on center of inertia over rough terrain,” �� , vol. 2, no. 1, p. 100031, 2022

2022
[23]

Momentum- aware trajectory optimization and control for agile quadrupedal locomotion,

Z. Zhou, B. Wingo, N. Boyd, S. Hutchinson, and Y . Zhao, “Momentum- aware trajectory optimization and control for agile quadrupedal locomotion,” �� , vol. 7, no. 3, pp. 7755–7762, 2022

2022
[24]

A linearization of centroidal dynamics for the model-predictive control of quadruped robots,

W. Chi, X. Jiang, and Y . Zheng, “A linearization of centroidal dynamics for the model-predictive control of quadruped robots,” in �� . IEEE, 2022, pp. 4656–4663

2022
[25]

Rloc: Terrain-aware legged locomotion using reinforcement learning and optimal control,

S. Gangapurwala, M. Geisert, R. Orsolino, M. Fallon, and I. Havoutis, “Rloc: Terrain-aware legged locomotion using reinforcement learning and optimal control,” �� , vol. 38, no. 5, pp. 2908–2927, 2022

2022
[26]

Learning agile and dynamic motor skills for legged robots. sci,

J. Hwangbo, J. Lee, A. Dosovitskiy, D. Bellicoso, V . Tsounis, V . Koltun, and M. Hutter, “Learning agile and dynamic motor skills for legged robots. sci,” ��, vol. 4, p. 26, 2019

2019
[27]

Learning multiple gaits within latent space for quadruped robots,

J. Wu, Y . Xue, and C. Qi, “Learning multiple gaits within latent space for quadruped robots,” �� , 2023

2023
[28]

Moral: Learning morphologically adaptive locomotion controller for quadrupedal robots on challenging terrains,

Z. Luo, Y . Dong, X. Li, R. Huang, Z. Shu, E. Xiao, and P. Lu, “Moral: Learning morphologically adaptive locomotion controller for quadrupedal robots on challenging terrains,” �� , 2024

2024
[29]

Synloco: Synthesizing central pattern generator and reinforcement learning for quadruped locomotion,

X. Zhang, Z. Xiao, Q. Zhang, and W. Pan, “Synloco: Synthesizing central pattern generator and reinforcement learning for quadruped locomotion,” in �� . IEEE, 2024, pp. 2640–2645

2024
[30]

Learning-based navigation and collision avoidance through reinforcement for uavs,

R. Azzam, M. Chehadeh, O. A. Hay, M. A. Humais, I. Boiko, and Y . Zweiri, “Learning-based navigation and collision avoidance through reinforcement for uavs,” �� , vol. 60, no. 3, pp. 2614–2628, 2023

2023
[31]

Flying in highly dynamic environments with end-to-end learning approach,

X. Fan, M. Lu, B. Xu, and P. Lu, “Flying in highly dynamic environments with end-to-end learning approach,” �� , vol. 10, no. 4, p. 3851–3858, Apr. 2025. [Online]. Available: http://dx.doi.org/10.1109/LRA.2025.3547306

2025
[32]

Review of wheeled mobile robot collision avoidance under unknown environment,

Y . Wang, X. Li, J. Zhang, S. Li, Z. Xu, and X. Zhou, “Review of wheeled mobile robot collision avoidance under unknown environment,” �� , vol. 104, no. 3, p. 00368504211037771, 2021

2021
[33]

Armor: Egocentric perception for humanoid robot collision avoidance and motion planning,

D. Kim, M. Srouji, C. Chen, and J. Zhang, “Armor: Egocentric perception for humanoid robot collision avoidance and motion planning,” �� , 2024

2024
[34]

Mpc-based optimization design for 3d collision avoidance of a mobile manipulator based-on obstacle velocity estimation,

K.-T. Song and C.-H. Lin, “Mpc-based optimization design for 3d collision avoidance of a mobile manipulator based-on obstacle velocity estimation,” in �� . IEEE, 2024, pp. 1–6

2024
[35]

Catch it! learning to catch in flight with mobile dexterous hands,

Y . Zhang, T. Liang, Z. Chen, Y . Ze, and H. Xu, “Catch it! learning to catch in flight with mobile dexterous hands,” �� , 2024

2024
[36]

Diversity is all you need: Learning skills without a reward function,

B. Eysenbach, A. Gupta, J. Ibarz, and S. Levine, “Diversity is all you need: Learning skills without a reward function,” �� , 2018

2018
[37]

Learning to walk in minutes using massively parallel deep reinforcement learning,

N. Rudin, D. Hoeller, P. Reist, and M. Hutter, “Learning to walk in minutes using massively parallel deep reinforcement learning,” in �� . PMLR, 2022, pp. 91–100

2022
[38]

Learning quadrupedal high-speed running on uneven terrain,

X. Han and M. Zhao, “Learning quadrupedal high-speed running on uneven terrain,” ��, vol. 9, no. 1, p. 37, 2024

2024
[39]

Zsl-rppo: Zero-shot learning for quadrupedal locomotion in challenging terrains using recurrent proximal policy optimization,

Y . Zhao, T. Wu, Y . Zhu, X. Lu, J. Wang, H. Bou-Ammar, X. Zhang, and P. Du, “Zsl-rppo: Zero-shot learning for quadrupedal locomotion in challenging terrains using recurrent proximal policy optimization,” �� , 2024

2024
[40]

A comparison of ppo, td3 and sac reinforcement algorithms for quadruped walking gait generation,

J. W. Mock and S. S. Muknahallipatna, “A comparison of ppo, td3 and sac reinforcement algorithms for quadruped walking gait generation,” �� , vol. 15, no. 1, pp. 36–56, 2023. APPENDIX I RL P OLICY TRAINING DETAILS We trained...

2023
[2025]

Available: https://arxiv.org/abs/2502.03132

[Online]. Available: https://arxiv.org/abs/2502.03132

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.

[1] [1]

Re- thinking robustness assessment: Adversarial attacks on learning-based quadrupedal locomotion controllers,

F. Shi, C. Zhang, T. Miki, J. Lee, M. Hutter, and S. Coros, “Re- thinking robustness assessment: Adversarial attacks on learning-based quadrupedal locomotion controllers,” �� , 2024

work page 2024

[2] [2]

Mobile robot collision avoidance based on deep reinforcement learning with motion constraints,

Y . Tao, M. Li, X. Cao, and P. Lu, “Mobile robot collision avoidance based on deep reinforcement learning with motion constraints,” �� , 2024

work page 2024

[3] [3]

Spark: A modular benchmark for humanoid robot safety,

Y . Sun, R. Chen, K. S. Yun, Y . Fang, S. Jung, F. Li, B. Li, W. Zhao, and C. Liu, “Spark: A modular benchmark for humanoid robot safety,”

work page

[4] [4]

Dynamic obstacle avoidance for quadrotors with event cameras,

D. Falanga, K. Kleber, and D. Scaramuzza, “Dynamic obstacle avoidance for quadrotors with event cameras,” �� , vol. 5, no. 40, p. eaaz9712, 2020

work page 2020

[5] [5]

Learning a state representation and navigation in cluttered and dynamic envi- ronments,

D. Hoeller, L. Wellhausen, F. Farshidian, and M. Hutter, “Learning a state representation and navigation in cluttered and dynamic envi- ronments,” �� , vol. 6, no. 3, pp. 5081–5088, 2021

work page 2021

[6] [6]

Robust autonomous navigation of a small-scale quadruped robot in real-world environments,

T. Dudzik, M. Chignoli, G. Bledt, B. Lim, A. Miller, D. Kim, and S. Kim, “Robust autonomous navigation of a small-scale quadruped robot in real-world environments,” in �� . IEEE, 2020, pp. 3664–3671

work page 2020

[7] [7]

Learning vision-guided quadrupedal locomotion end-to-end with cross-modal transformers,

R. Yang, M. Zhang, N. Hansen, H. Xu, and X. Wang, “Learning vision-guided quadrupedal locomotion end-to-end with cross-modal transformers,” �� , 2021. �� Fig. 8: REBot system real-robot demonstrations on Unitree Go2 Robot (See video). (a) the robot is poked from different directions using a stick; (b) a ball is launche...

work page 2021

[8] [8]

Creating a dynamic quadrupedal robotic goalkeeper with reinforcement learning,

X. Huang, Z. Li, Y . Xiang, Y . Ni, Y . Chi, Y . Li, L. Yang, X. B. Peng, and K. Sreenath, “Creating a dynamic quadrupedal robotic goalkeeper with reinforcement learning,” in �� . IEEE, 2023, pp. 2715–2722

work page 2023