Pith. sign in

REVIEW 3 major objections 7 minor 34 references

Versatile Locomotion Skills for Hexapod Robots

T0 review · 3 major / 7 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read A simulation-only training pipeline gives a $600 hexapod robot vision-based skills to climb stairs, avoid obstacles, and squeeze under low objects in the physical world.

desk verdict Solid systems demo of vision-based hexapod skills, but the stair metric is average steps completed, not per-trial success, and the abstract overstates it. read the letter →

arxiv 2412.10628 v1 pith:45I6IEEI submitted 2024-12-14 cs.RO

Tomson Qu , Dichen Li , Avideh Zakhor , Wenhao Yu , Tingnan Zhang This is my paper · ORCID
classification cs.RO
keywords hexapodlocomotionsim-to-realtransferteacher-studentdistillationreinforcementlearningstairclimbingobstacleavoidancesqueezingunderobstaclesvisual-inertialodometry
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 sets out to show that a $600 hexapod robot, equipped only with a depth camera and visual-inertial odometry (VIO), can learn to climb up and down staircases, avoid obstacles, and squeeze under low objects without any real-time joint feedback and without any real-world training. The key move is a two-stage teacher-student pipeline: first a reinforcement-learning teacher is trained in simulation with privileged information, then a student policy is distilled to act from egocentric depth images and body pose only. Physical experiments report stair completions of 6.0 to 7.6 out of 7 to 8 steps, obstacle-avoidance success in 7 to 9 of 10 trials, and squeeze success rates of 85 to 100 percent. If the result holds, perceptive legged locomotion becomes available on sub-thousand-dollar hardware.

What carries the argument

The central mechanism is a two-stage teacher-student distillation trained entirely in simulation. In phase 1, a reinforcement-learning teacher observes privileged information—a terrain elevation map plus joint feedback—and is rewarded with task-specific terms, including a distance-weighted obstacle penalty for avoidance and a body-to-obstacle/ground distance reward for squeezing; curriculum training gradually raises stair riser height and difficulty, obstacle density, and squeeze demands. In phase 2, supervised learning distills the teacher into a student that maps a 320×240 egocentric depth image and a VIO pose estimate directly to 18 joint angles, so the deployed robot needs no real-time joint-state feedback.

What would settle it

Record the robot's 18 joint-angle commands and VIO pose during a stair climb on a fresh 12-volt battery and replay the same depth-image inputs in the simulator: if the real joint trajectories diverge substantially from the simulated rollout, or if the stair success rate over ten trials falls well below the reported 6.0 to 7.6 average steps, the zero-shot sim-to-real claim is falsified.

Watch

Extended reading notes

Core claim

The central discovery is that a vision-based locomotion policy for a low-cost hexapod can be trained entirely in simulation and then deployed zero-shot on the physical robot without real-time joint-state feedback. The teacher policy, trained with privileged terrain elevation information and joint feedback, is distilled into a student policy whose only observations are a 320×240 egocentric depth image and a VIO pose estimate; the student directly outputs 18 joint angles. The paper demonstrates this recipe across three distinct tasks, with task-specific terrain construction, reward shaping, and camera orientation, and reports that the resulting behaviors generalize to unseen staircases, irregular obstacles, and tunnels of different lengths.

Load-bearing premise

The load-bearing premise is that the simulation's physics—especially the robot's servo-driven joints and contact dynamics—matches the real $600 robot closely enough that a policy trained only in simulated data transfers without any real-world fine-tuning.

Editorial extensions

If this is right

  • If the result holds, perception-driven legged locomotion no longer requires expensive joint encoders and real-time control loops; a depth camera, VIO, and roughly $600 of hardware suffice for several practical tasks.
  • The same two-stage recipe can be applied to new tasks by changing only the terrain, reward terms, and camera angle, which is the paper's stated path toward a single universal policy.
  • The squeezing experiments show the controller can hold a crouched posture through a 129-centimeter tunnel and re-raise afterward, indicating the learned behavior is a sustained skill rather than a short transient.
  • The avoidance policy's global-y deviation reward makes the robot return to its original heading after passing an obstacle, a behavior the manufacturer's baseline ultrasound avoidance does not exhibit.
  • The stair-climbing policy transfers across three real staircases with different riser and tread dimensions, suggesting the learned controller is not overfit to one geometry.

Reading between the lines

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

  • Beyond the paper: the three task policies require different camera pitches—30 degrees downward for stairs and avoidance, level for squeezing—so a single combined policy will likely need an actuated or switchable camera mount; the paper names this as future work.
  • Beyond the paper: the reported sensitivity to battery voltage suggests that modeling servo voltage sag or randomized actuation delay in simulation could further close the sim-to-real gap, and that the measured success rates may be optimistic at nominal conditions only.
  • Beyond the paper: the reward-shaping ideas—distance-weighted obstacle maps and body-clearance rewards—are not hexapod-specific and could transfer to other legged robots as generic shaping terms.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 7 minor

Summary. The paper presents a two-stage teacher-student reinforcement learning method for training a low-cost hexapod robot (SpiderPi, ~$600) to perform three locomotive skills: stair climbing, obstacle avoidance, and squeezing under low obstacles. The teacher policy is trained in Isaac Gym with privileged observations (height maps, joint states) and is distilled into a student policy that uses only egocentric depth images and VIO pose. The authors deploy the student policy zero-shot on the physical robot and report physical experiments on three campus staircases, four obstacle-avoidance scenarios (including stationary and moving persons), and four squeezing configurations. Reported success rates range from 7/10 to 9/10 for avoidance and 85% to 100% for squeezing, while the stair experiments report average completed steps between 6.0/7.0 and 7.6/8.0 over 10 trials.

Significance. If the results are taken at face value, the paper demonstrates that a sub-thousand-dollar hexapod can use learned vision-based policies, trained entirely in simulation, to perform multiple real-world locomotion skills without real-time joint feedback. The physical experiments on real staircases, cluttered paths, and restrictive tunnels are the paper's main strength, and the teacher-student architecture is a sensible approach for handling the absence of joint-state feedback. The contribution is incremental over the authors' previous joist-climbing work, but extends the approach to three new skills and provides useful empirical evidence of zero-shot transfer. The paper is weakened by the misleading metric used for stair climbing and by the absence of uncertainty quantification across all experiments.

major comments (3)
  1. [Section IV.A, Table III, Abstract] The abstract and introduction state that the robot achieves 'high success rates' on stair climbing, but Table III does not report success rates. It reports the average number of stairs completed over 10 trials (e.g., 6.0/7.0 for Cory Hall). A trial in which the robot completes 6 of 7 steps and then falls or gets stuck is counted as 6/7, not as a failure, so the numbers cannot be interpreted as the fraction of trials that successfully climbed the entire staircase. This overstates the central claim. Please report the per-trial success count (e.g., number of fully successful trials out of 10) together with the distribution of completed steps, or at least the number of trials that reached the top, and adjust the abstract accordingly.
  2. [Section IV, Tables III-V] The physical experiments are presented without any measure of uncertainty. With only 10 trials per condition in Tables III and IV, and 10–20 in Table V, the differences between conditions (e.g., 7/10 vs. 9/10 in Table IV) are not statistically meaningful, and the average values in Table III are highly sensitive to the underlying trial distribution. The authors should provide per-trial data or at least binomial confidence intervals, and should avoid comparative statements such as 'Soda Hall has the highest success rate' (Section IV.A) unless they are supported by a statistical test.
  3. [Section III, Section IV.C] The central claim of zero-shot sim-to-real transfer is supported by the physical experiments, but the manuscript provides no quantitative description of the simulation-to-reality match: no comparison of simulated and real body trajectories, no actuator latency or servo model, and no details of the domain randomization presumably used in training (the related work cites domain randomization, but Section III does not specify any). The footnote in Section IV.C on battery voltage sensitivity indicates that unmodeled dynamics can degrade performance. To make the zero-shot claim reproducible and to characterize its limits, the authors should report the simulation parameters, randomization ranges, and any hardware-specific modeling (e.g., servo bandwidth, communication latency).
minor comments (7)
  1. [Section IV.A] The text says 'In Figure 5, Cory Hall and Soda Hall each had 7 steps' but the staircases are shown in Figure 6; the cross-reference should be corrected.
  2. [Table III] The name 'Sudardja' in Table III is spelled 'Sutardjai' in the text; use the correct spelling consistently.
  3. [Section III.C] The sentence 'Since Isaac Gym does not support floating terrains and requires everything to be grounded, we modified the code to construct terrains on vertices "in the air"' is awkward and potentially misleading; rephrase to clarify whether the modification was to Isaac Gym's terrain generation or to the height-map representation.
  4. [Section IV] The paper refers to 'video link' and 'this video link' without providing actual URLs or a supplementary material section; without these, the qualitative claims (e.g., the 15-step staircase with a landing, the baseline comparison in Section IV.A) cannot be verified.
  5. [References] Reference [1] is incomplete: it lists authors and a URL but no title or venue; this should be fixed.
  6. [Section IV.C] The battery-voltage sensitivity mentioned in the footnote is an important limitation that affects all three tasks; consider moving it into Section V (Limitations) so it is not easily overlooked.
  7. [Section IV.A] The comparison against the baseline 'walk' policy on stairs is described only qualitatively ('crashing down the stairs'); if this baseline is intended as a point of comparison, include at least a summary statistic to make the contrast explicit.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity found; the central sim-to-real claim is verified by physical experiments rather than derived from fitted inputs.

full rationale

The paper's central claim is empirical: a teacher-student policy trained exclusively in Isaac Gym transfers zero-shot to a $600 hexapod for stair climbing, obstacle avoidance, and squeezing, as evaluated in physical trials (Section IV, Tables III-V). This is not a derivation in which a predicted quantity is algebraically equivalent to an input by construction. The reward functions and terrain curricula in Section III shape the learned behavior, and some success metrics overlap with reward objectives (e.g., collision-free traversal mirrors the collision penalty), but this is objective alignment, not circular reasoning: the physical trials are external to the training loop and are not used to fit any parameter. The self-citations to the authors' prior work [2] define the teacher-student framework and the top twelve reward terms, but the present paper's physical transfer results stand independently of those citations, and no uniqueness theorem or unverified premise is imported exclusively from [2]. The stair metric in Table III reports average stairs completed per trial rather than per-trial success; this is a reporting and interpretability concern, not a circularity, and belongs in the correctness pass. No specific reduction of any claimed outcome to its own input was found, so the circularity score is 0.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The paper introduces no invented physical entities, forces, or conserved quantities. The central result rests on hand-tuned reward weights, per-task camera angles and height-map geometries, curriculum schedules, and five domain assumptions about simulator fidelity, perception, actuation, reward encoding, and the privileged representation. The modified floating-terrain support in Isaac Gym is a software change, not an entity.

free parameters (7)
  • Per-task reward weights (Table I) = e.g., forward velocity weight 12, lateral velocity weight -11, global-y deviation -12 for stairs, collision penalty…
    Hand-tuned to shape behavior for each task; no sensitivity analysis is provided.
  • Camera angle per task (Table II) = 30 degrees for stairs and avoidance, 0 degrees for squeezing
    Empirically found optimal; the paper notes squeezing requires a level camera to see the tunnel.
  • Height map size and location per task (Table II) = 0.6 x 0.8 m at 0.3 m for stairs, 0.6 x 1.0 m at 0.6 m for avoidance, 0.6 x 0.8 m at 0 m for squeezing
    Empirically tuned; avoidance needs a larger, farther map.
  • Stair curriculum schedule = riser heights 4.5 to 18 cm, tread depths 30 to 18 cm
    Curriculum levels hand-set to increase difficulty gradually.
  • Obstacle density curriculum = density = (2 * level / total_levels) * density_final
    Formula chosen to avoid early local optima from dense obstacles.
  • Squeezing obstacle heights = 37, 35, 33, 31 cm
    Curriculum levels for decreasing clearance between obstacle and ground.
  • Weight vectors w1, w2, w3, w4 shapes = triangular and ramp profiles over height-map cells
    Hand-designed to emphasize nearby and centered obstacles in the distance-to-obstacle rewards.
assumptions (5)
  • domain assumption The Isaac Gym simulation, including URDF, contacts, and actuation, is faithful enough to the physical SpiderPi for zero-shot sim-to-real transfer.
    Invoked in Section III (training) and Section IV (deployment); no quantitative sim-to-real comparison is provided.
  • domain assumption The T265 VIO pose and L515 depth image provide sufficient observation for the student policy without joint feedback.
    Stated in the abstract and Section I; the policy outputs joint angles directly from these observations.
  • domain assumption Position-controlled servos reach commanded angles accurately under load without closed-loop feedback.
    Implied by the low-cost hardware design in Section I; the battery-voltage sensitivity in Section IV suggests this is fragile.
  • domain assumption The hand-designed reward functions encode the desired skills, including stairs, avoidance, and squeezing.
    Section III defines task-specific rewards; the policy optimizes them, so behavior is shaped by these choices.
  • domain assumption The privileged height-map representation is sufficient for learning the skills in phase 1.
    Section III uses height maps of different sizes and locations per task; the student must mimic this from depth images.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Versatile Locomotion Skills for Hexapod Robots." pith.science (2026). https://pith.science/paper/45I6IEEI

@misc{pith2026241210628,
  author       = {Pith},
  title        = {Pith review of: Versatile Locomotion Skills for Hexapod Robots},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/45I6IEEI}},
  note         = {Machine review of arXiv:2412.10628}
}
read the original abstract

Hexapod robots are potentially suitable for carrying out tasks in cluttered environments since they are stable, compact, and light weight. They also have multi-joint legs and variable height bodies that make them good candidates for tasks such as stairs climbing and squeezing under objects in a typical home environment or an attic. Expanding on our previous work on joist climbing in attics, we train a legged hexapod equipped with a depth camera and visual inertial odometry (VIO) to perform three tasks: climbing stairs, avoiding obstacles, and squeezing under obstacles such as a table. Our policies are trained with simulation data only and can be deployed on lowcost hardware not requiring real-time joint state feedback. We train our model in a teacher-student model with 2 phases: In phase 1, we use reinforcement learning with access to privileged information such as height maps and joint feedback. In phase 2, we use supervised learning to distill the model into one with access to only onboard observations, consisting of egocentric depth images and robot pose captured by a tracking VIO camera. By manipulating available privileged information, constructing simulation terrains, and refining reward functions during phase 1 training, we are able to train the robots with skills that are robust in non-ideal physical environments. We demonstrate successful sim-to-real transfer and achieve high success rates across all three tasks in physical experiments.

Figures

Figures reproduced from arXiv: 2412.10628 by the authors.

Figure 1
Figure 1. Physical and URDF of the robot. (a) The hexapod robot standing at the reset position - roughly 37 cm tall. (b) The URDF of the hexapod. business, it is important for legged locomotion controllers to work with low-cost hardware. However, most existing legged locomotion systems require high-end robots capable of real-time sensing of joint states, which could ultimately result in expensive hardware on the order of thou… view at source ↗
Figure 2
Figure 2. High-level overview of training methodology [2] [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 5
Figure 5. Physical design of a squeezing environment. [PITH_FULL_IMAGE:figures/full_fig_p004_5.png] view at source ↗
Figures from the paper (6 more)
Figure 6
Figure 6. Figure 6: Experiments Scene for stairs climbing. The first number is the riser [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 7
Figure 7. Figure 7: Obstacle Avoidance Environment. control and falling down. We present a video link showing the robot climbing two sets of stairs separated by a platform, with 8 and 7 steps in the first and second sets respectively. As seen in the video, the robot is able to traverse in…
Figure 8
Figure 8. Figure 8: Experiments Scene for Squeezing. (a) Metal rod as obstacle without tunnel. (b) Lengthy tunnel. (c) Paper block as obstacle without tunnel. (d) Paper block as obstacle with tunnel. the left and right depending on its distance to each side of the obstacle. This indicates…
Figure 9
Figure 9. Figure 9: Squeezing under two consecutive obstacles. [PITH_FULL_IMAGE:figures/full_fig_p007_9.png]
Figure 10
Figure 10. Figure 10: Height of robot body vs. rollout step in Isaac Gym. [PITH_FULL_IMAGE:figures/full_fig_p007_10.png]
Figure 8
Figure 8. Figure 8: As seen, the success rate is high2 . While the robot can instantaneously squeeze as low as 31.75 cm to go under a thin metal rod, it cannot sustain that height for an extended distance of say 129 cm without hitting the tunnel roof. However, it can sustain the squeeze p…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 25 canonical work pages

  1. [1]

    Zakhor, E

    A. Zakhor, E. Lathrop, A. Austin, J. Zang, M. Tang, A. Roychowdry, W. Naing, M. K. Beauden, and Y . Li. [Online]. Available: https://www-video.eecs.berkeley.edu/papers/avz/revised-icra-2022.pdf

    work page 2022

  2. [2]

    Per- ceptive hexapod legged locomotion for climbing joist environments,

    Z. Zang, M. Kawawa-Beaudan, W. Yu, T. Zhang, and A. Zakhor, “Per- ceptive hexapod legged locomotion for climbing joist environments,” in 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2023, pp. 2738–2745

    work page 2023

  3. [3]

    Sim-to-real: Learning agile locomotion for quadruped robots,

    J. Tan, T. Zhang, E. Coumans, A. Iscen, Y . Bai, D. Hafner, S. Bo- hez, and V . Vanhoucke, “Sim-to-real: Learning agile locomotion for quadruped robots,” 2018

    2018

  4. [4]

    Legged locomotion in challenging terrains using egocentric vision,

    A. Agarwal, A. Kumar, J. Malik, and D. Pathak, “Legged locomotion in challenging terrains using egocentric vision,” 2022

    work page 2022

  5. [5]

    Oscar pistorius: The blade runner makes olympic history,

    K. Paramaguru, “Oscar pistorius: The blade runner makes olympic history,” Jul 2012. [Online]. Available: https://olympics.time.com/ 2012/07/05/oscar-pistorius-the-blade-runner-makes-olympic-history/

    work page 2012

  6. [6]

    Adaptive Locomotion Control of a Hexapod Robot via Bio-Inspired Learning,

    W. Ouyang, H. Chi, J. Pang, W. Liang, and Q. Ren, “Adaptive Locomotion Control of a Hexapod Robot via Bio-Inspired Learning,” Frontiers in Neurorobotics , vol. 15, p. 627157, Jan. 2021. [Online]. Available: https://www.frontiersin.org/articles/10.3389/fnbot. 2021.627157/full

    arXiv 2021

  7. [7]

    How to train your robot with deep reinforcement learning: lessons we have learned,

    J. Ibarz, J. Tan, C. Finn, M. Kalakrishnan, P. Pastor, and S. Levine, “How to train your robot with deep reinforcement learning: lessons we have learned,” The International Journal of Robotics Research , vol. 40, no. 4-5, pp. 698–721, Apr. 2021. [Online]. Available: http://journals.sagepub.com/doi/10.1177/0278364920987859

    doi:10.1177/0278364920987859 2021

  8. [8]

    Sim-to-real via latent prediction: Transferring visual non-prehensile manipulation policies,

    C. Rizzardo, F. Chen, and D. Caldwell, “Sim-to-real via latent prediction: Transferring visual non-prehensile manipulation policies,” Frontiers in Robotics and AI , vol. 9, p. 1067502, Jan. 2023. [Online]. Available: https://www.frontiersin.org/articles/10.3389/frobt. 2022.1067502/full

    arXiv 2023

Show all 34 references
  1. [9]

    Learning for a Robot: Deep Reinforcement Learning, Imitation Learning, Transfer Learning,

    J. Hua, L. Zeng, G. Li, and Z. Ju, “Learning for a Robot: Deep Reinforcement Learning, Imitation Learning, Transfer Learning,” Sensors, vol. 21, no. 4, p. 1278, Feb. 2021. [Online]. Available: https://www.mdpi.com/1424-8220/21/4/1278

    2021

  2. [10]

    Sim-to-Real: Learning Agile Locomotion For Quadruped Robots,

    J. Tan, T. Zhang, E. Coumans, A. Iscen, Y . Bai, D. Hafner, S. Bohez, and V . Vanhoucke, “Sim-to-Real: Learning Agile Locomotion For Quadruped Robots,” May 2018, arXiv:1804.10332 [cs]. [Online]. Available: http://arxiv.org/abs/1804.10332

    2018 arXiv

  3. [11]

    Sim-to-Real Transfer in Deep Reinforcement Learning for Robotics: a Survey,

    W. Zhao, J. P. Queralta, and T. Westerlund, “Sim-to-Real Transfer in Deep Reinforcement Learning for Robotics: a Survey,” in 2020 IEEE Symposium Series on Computational Intelligence (SSCI) . Canberra, ACT, Australia: IEEE, Dec. 2020, pp. 737–744. [Online]. Available: https://i...

    2020

  4. [12]

    DROPO: Sim-to-real transfer with offline domain randomization,

    G. Tiboni, K. Arndt, and V . Kyrki, “DROPO: Sim-to-real transfer with offline domain randomization,” Robotics and Autonomous Systems, vol. 166, p. 104432, Aug. 2023. [Online]. Available: https://linkinghub.elsevier.com/retrieve/pii/S0921889023000714

    2023

  5. [13]

    Deep reinforcement learning for modeling human locomotion control in neuromechanical simulation,

    S. Song, L. Kidzinski, X. B. Peng, C. Ong, J. Hicks, S. Levine, C. G. Atkeson, and S. L. Delp, “Deep reinforcement learning for modeling human locomotion control in neuromechanical simulation,” Journal of NeuroEngineering and Rehabilitation , vol. 18, no. 1, p. 126, Aug

  6. [14]

    Deep Reinforcement Learning for Snake Robot Locomotion,

    J. Shi, T. Dear, and S. D. Kelly, “Deep Reinforcement Learning for Snake Robot Locomotion,” IFAC-PapersOnLine, vol. 53, no. 2, pp. 9688–9695, 2020. [Online]. Available: https://linkinghub.elsevier. com/retrieve/pii/S2405896320333772

    2020

  7. [15]

    Model-free reinforcement learning for robust locomotion using demonstrations from trajectory optimization,

    M. Bogdanovic, M. Khadiv, and L. Righetti, “Model-free reinforcement learning for robust locomotion using demonstrations from trajectory optimization,” Frontiers in Robotics and AI, vol. 9, p. 854212, Aug. 2022. [Online]. Available: https://www.frontiersin.org/articles/10.3389...

    2022

  8. [16]

    Learning and Reusing Quadruped Robot Movement Skills from Biological Dogs for Higher-Level Tasks,

    Q. Wan, A. Luo, Y . Meng, C. Zhang, W. Chi, S. Zhang, Y . Liu, Q. Zhu, S. Kong, and J. Yu, “Learning and Reusing Quadruped Robot Movement Skills from Biological Dogs for Higher-Level Tasks,” Sensors, vol. 24, no. 1, p. 28, Dec. 2023. [Online]. Available: https://www.mdpi.com/1...

    2023

  9. [17]

    Robot Parkour Learning,

    Z. Zhuang, Z. Fu, J. Wang, C. Atkeson, S. Schwertfeger, C. Finn, and H. Zhao, “Robot Parkour Learning,” Sep. 2023, arXiv:2309.05665 [cs]. [Online]. Available: http://arxiv.org/abs/2309.05665

    2023 arXiv

  10. [18]

    ViNL: Visual Navigation and Locomotion Over Obstacles,

    S. Kareer, N. Yokoyama, D. Batra, S. Ha, and J. Truong, “ViNL: Visual Navigation and Locomotion Over Obstacles,” Oct. 2023, arXiv:2210.14791 [cs]. [Online]. Available: http://arxiv.org/abs/2210. 14791

    2023 arXiv

  11. [19]

    Visual-Locomotion: Learning to Walk on Complex Terrains with Vision

    W. Yu, D. Jain, A. Escontrela, A. Iscen, P. Xu, E. Coumans, S. Ha, J. Tan, and T. Zhang, “Visual-Locomotion: Learning to Walk on Complex Terrains with Vision.”

  12. [20]

    The dynamic window approach to collision avoidance,

    D. Fox, W. Burgard, and S. Thrun, “The dynamic window approach to collision avoidance,” IEEE Robotics Automation Magazine , vol. 4, no. 1, pp. 23–33, 1997

    1997

  13. [21]

    Reinforcement learning-based dynamic obstacle avoidance and integration of path planning,

    J. Choi, G. Lee, and C. Lee, “Reinforcement learning-based dynamic obstacle avoidance and integration of path planning,” Intelligent Service Robotics , vol. 14, no. 5, pp. 663–677, Nov. 2021. [Online]. Available: https://link.springer.com/10.1007/s11370-021-00387-2

    2021 doi

  14. [22]

    Enhanced method for reinforcement learning based dynamic obstacle avoidance by assessment of collision risk,

    F. Hart and O. Okhrin, “Enhanced method for reinforcement learning based dynamic obstacle avoidance by assessment of collision risk,” Neurocomputing, vol. 568, p. 127097, Feb. 2024. [Online]. Available: https://linkinghub.elsevier.com/retrieve/pii/S0925231223012201

    2024

  15. [23]

    Learning quadrupedal locomotion over challenging terrain,

    J. Lee, J. Hwangbo, L. Wellhausen, V . Koltun, and M. Hutter, “Learning quadrupedal locomotion over challenging terrain,” Science Robotics, vol. 5, no. 47, Oct. 2020. [Online]. Available: http: //dx.doi.org/10.1126/scirobotics.abc5986

    2020 doi

  16. [24]

    Rma: Rapid motor adaptation for legged robots,

    A. Kumar, Z. Fu, D. Pathak, and J. Malik, “Rma: Rapid motor adaptation for legged robots,” 2021

    2021

  17. [25]

    Reinforcement Learning-Based Reactive Obstacle Avoidance Method for Redundant Manipulators,

    Y . Shen, Q. Jia, Z. Huang, R. Wang, J. Fei, and G. Chen, “Reinforcement Learning-Based Reactive Obstacle Avoidance Method for Redundant Manipulators,” Entropy, vol. 24, no. 2, p. 279, Feb

  18. [26]

    An obstacle avoidance- specific reinforcement learning method based on fuzzy attention mechanism and heterogeneous graph neural networks,

    F. Zhang, C. Xuan, and H.-K. Lam, “An obstacle avoidance- specific reinforcement learning method based on fuzzy attention mechanism and heterogeneous graph neural networks,” Engineering Applications of Artificial Intelligence , vol. 130, p. 107764, Apr

  19. [27]

    Overcoming Obstacles With a Reconfigurable Robot Using Reinforcement Learning,

    L. Yehezkel, S. Berman, and D. Zarrouk, “Overcoming Obstacles With a Reconfigurable Robot Using Reinforcement Learning,” IEEE Access, vol. 8, pp. 217 541–217 553, 2020. [Online]. Available: https://ieeexplore.ieee.org/document/9272763/

    2020

  20. [28]

    Virtual Robot Climbing using Reinforcement Learning,

    U. Garg, “Virtual Robot Climbing using Reinforcement Learning,” Master of Science, San Jose State University, San Jose, CA, USA, Dec. 2018. [Online]. Available: https://scholarworks.sjsu.edu/ etd projects/658

    2018

  21. [29]

    Blind Bipedal Stair Traversal via Sim-to-Real Reinforcement Learning,

    J. Siekmann, K. Green, J. Warila, A. Fern, and J. Hurst, “Blind Bipedal Stair Traversal via Sim-to-Real Reinforcement Learning,” in Robotics: Science and Systems XVII . Robotics: Science and Systems Foundation, Jul. 2021. [Online]. Available: http://www.roboticsproceedings.org...

    2021

  22. [30]

    Reinforcement Learning Based, Staircase Negotiation Learning: Simulation and Transfer to Reality for Articulated Tracked Robots,

    A. Mitriakov, P. Papadakis, J. Kerdreux, and S. Garlatti, “Reinforcement Learning Based, Staircase Negotiation Learning: Simulation and Transfer to Reality for Articulated Tracked Robots,” IEEE Robotics Automation Magazine , vol. 28, no. 4, pp. 10–20, Dec. 2021. [Online]. Avai...

    2021

  23. [31]

    Staircase Traversal via Reinforcement Learning for Active Reconfiguration of Assistive Robots,

    A. Mitriakov, P. Papadakis, S. Mai Nguyen, and S. Garlatti, “Staircase Traversal via Reinforcement Learning for Active Reconfiguration of Assistive Robots,” in 2020 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE) . Glasgow, United Kingdom: IEEE, Jul. 2020, pp. 1–8. ...

    2020

  24. [2021]

    Available: https://jneuroengrehab.biomedcentral.com/ articles/10.1186/s12984-021-00919-y

    [Online]. Available: https://jneuroengrehab.biomedcentral.com/ articles/10.1186/s12984-021-00919-y

    doi

  25. [2022]

    Available: https://www.mdpi.com/1099-4300/24/2/279

    [Online]. Available: https://www.mdpi.com/1099-4300/24/2/279

  26. [2024]

    Available: https://linkinghub.elsevier.com/retrieve/pii/ S0952197623019486

    [Online]. Available: https://linkinghub.elsevier.com/retrieve/pii/ S0952197623019486

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.