Pith. sign in

REVIEW 4 major objections 6 minor 4 cited by

Omni-Perception: Omnidirectional Collision Avoidance for Legged Locomotion in Dynamic Environments

T0 review · 4 major / 6 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 Omni-Perception claims the first end-to-end legged locomotion policy trained directly on raw LiDAR point clouds, achieving omnidirectional collision avoidance in dynamic 3D scenes.

desk verdict First raw-LiDAR end-to-end legged locomotion policy with real-world promise, but sim-to-real fidelity is asserted more than measured. read the letter →

arxiv 2505.19214 v2 pith:X5ES7BET submitted 2025-05-25 cs.RO

Zifan Wang , Teli Ma , Yufei Jia , Xun Yang , Jiaming Zhou , Wenlong Ouyang , Qiang Zhang , Junwei Liang This is my paper
classification cs.RO
keywords LeggedRobotLocomotionReinforcementLearningLiDARPerceptionCollisionAvoidanceSim-to-RealPointCloudPD-RiskNet
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 sets out to show that a quadruped can navigate dynamic, three-dimensional surroundings using only raw spatio-temporal LiDAR point clouds, learned end-to-end by reinforcement learning, without elevation maps or decoupled planning. It introduces PD-RiskNet, a hierarchical network that splits each scan into near and far regions, processes each with a recurrent unit, and feeds the result into a locomotion policy. To train this policy, the authors build a high-fidelity LiDAR simulator with realistic noise and non-repetitive scan patterns. On a real quadruped, the resulting policy avoids aerial obstacles and moving humans at success rates of 70 and 90 percent respectively, where the native system scores zero.

What carries the argument

The load-bearing component is PD-RiskNet, the Proximal-Distal Risk-Aware Hierarchical Network, a perception module that partitions each raw LiDAR frame into a proximal set of dense near-field points and a distal set of sparse far-field points. Each set is downsampled, ordered by spherical coordinates, and processed by its own gated recurrent unit, with the proximal unit receiving privileged height supervision during training. The resulting embeddings are concatenated with proprioceptive history and the velocity command, then passed to an MLP actor. A second mechanism is the sector-based avoidance velocity: the horizontal plane is divided into 36 angular sectors, each sector contributes a repulsive velocity whose magnitude decays exponentially with the distance of the nearest point in that sector, and the reward encourages tracking the command velocity plus this avoidance term.

What would settle it

Measure the perceptual gap by placing the same static scene in front of the real LiDAR and the simulator, computing the Chamfer distance between the two point clouds, then running the policy on a thin rod whose diameter is below the averaging kernel. If the Chamfer distance is large but the policy still transfers, or if the policy collides with thin rods that were clearly visible in simulation, the high-fidelity-transfer claim is weakened.

Watch

Extended reading notes

Core claim

The central claim is that end-to-end policies can consume raw LiDAR point clouds directly and achieve omnidirectional collision avoidance in dynamic 3D environments, something depth-camera pipelines and map-based methods do not deliver. The discovery is that a proximal-distal hierarchical processing scheme makes this tractable: dense near-field points are downsampled by farthest-point sampling, sparse far-field points by averaging, and each stream is fed to a separate gated recurrent unit trained with privileged height supervision. Combined with a sector-based avoidance velocity in the reward, this yields a reactive policy that tracks velocity commands while pushing away from nearby obstacles in all directions. The paper reports real-robot success rates of 100 percent on static obstacles, 70 percent on aerial obstacles, 83 percent on small obstacles, and 90 percent on moving humans, compared to zero for the native system on aerial and human obstacles.

Load-bearing premise

The whole sim-to-real transfer rests on the assumption that the simulated LiDAR scans faithfully reproduce the real sensor's noise, scan pattern, and self-occlusion; the paper supports this with a single qualitative side-by-side comparison and domain randomization, but offers no quantitative error metrics.

Editorial extensions

If this is right

  • If the claim holds, intermediate representations like elevation maps become optional for reactive legged navigation, since raw LiDAR provides lighting-invariant 3D awareness at control frequency.
  • The 70 percent aerial and 90 percent moving-human success rates imply that end-to-end LiDAR policies can handle threats that proprioception-only systems miss entirely.
  • The LiDAR simulator's speed, more than an order of magnitude faster than one commercial baseline on static scenes, makes large-scale parallel training of perception-locomotion policies practical.
  • The same policy transfers to a real robot without additional adaptation, indicating that zero-shot sim-to-real transfer is feasible for point-cloud-based legged locomotion.

Reading between the lines

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

  • The proximal-distal split is a generic way to handle non-uniform point density; it could be applied to other irregular point-cloud streams such as UAV navigation or manipulation where near-field detail matters more than far-field density.
  • A quantitative benchmark of LiDAR simulator fidelity, such as Chamfer distance or occupancy IoU between simulated and real scans, would let the community separate perception gap from control policy quality; the current paper only offers a qualitative comparison.
  • Extending the paper's failure analysis, adding semantic segmentation could address the dense-grass failure case, since geometric features alone misclassify vegetation as danger.
  • The avoidance-velocity reward is a reactive heuristic; a learned or predictive variant might improve behavior in corridors where the robot is forced to walk close to obstacles.
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

4 major / 6 minor

Summary. The paper proposes Omni-Perception, an end-to-end reinforcement learning framework for legged locomotion that consumes raw spatio-temporal LiDAR point clouds through a new perception module, PD-RiskNet, and outputs joint position targets. The method partitions each scan into proximal and distal clouds, processes them with FPS/GRU and average-downsampling/GRU pathways, and trains with a reward that combines velocity tracking with a hand-designed sector-based avoidance velocity and a LiDAR distance-maximization term. The authors also contribute a custom GPU-accelerated LiDAR simulator and report a simulation ablation plus real-robot trials across static, aerial, small, and moving-human obstacle scenarios.

Significance. If the sim-to-real transfer is robust, the paper would be a meaningful step toward using direct 3D LiDAR point clouds in learned legged locomotion, avoiding intermediate elevation maps and depth-camera limitations. The custom LiDAR simulation toolkit with fast parallel raycasting and cross-platform support is a practical contribution, as is the public code repository. The real-robot demonstrations show that a raw-point-cloud policy can produce reactive avoidance behaviors that the native Unitree system lacks. However, the quantitative evidence for the central zero-shot transfer claim is currently thinner than the narrative suggests: the LiDAR fidelity evaluation is qualitative, the real-world results are point estimates from 30 trials without statistical support, and the paper's own limitations predict the observed small-obstacle deficit. The significance is therefore real but conditional on stronger validation.

major comments (4)
  1. [4.1] The LiDAR simulator fidelity claim, which underpins the zero-shot sim-to-real transfer result, rests on a single qualitative side-by-side comparison in Figure 4. No quantitative metric is reported for per-ray range error, point-cloud recall, scan-pattern coverage, or self-occlusion between the physical Livox Mid-360 and the simulator. This is load-bearing because the policy does not consume raw scans directly: Section 3.3.1 applies FPS, average downsampling, and spherical sorting before the GRUs, so any bias in the simulated scan propagates into the policy observation. The paper's own Limitations section and Figure 8 state that averaging the distant point cloud destroys the features of very small or thin objects, and Table 4 shows exactly that class underperforming (83% vs. 100% for the non-perceptive Unitree system). Please add a quantitative sim-vs-real fidelity comparison or a closed-loop sim-vs-real benchmark, and discuss how the distal downsampling interacts with the fidelity claim.
  2. [4.4] The real-world success rates in Table 4 are point estimates from only 30 trials per scenario, with no confidence intervals, no statistical significance tests, and no description of the trial protocol (obstacle dimensions, approach speeds, initial configurations, or success criteria). For example, 21/30 successes for aerial obstacles has an exact 95% confidence interval of roughly 51% to 85%, which is much wider than the 70% point estimate suggests. Please report confidence intervals or additional repetitions, and specify the trial protocol so the reader can assess whether the differences are robust and reproducible.
  3. [4.3] In Table 3, the reported success rates and collision rates are not mutually exclusive and sum to more than 100% for several rows (e.g., FPS+MLP: 33.3% + 93.3% = 126.6%; Ours: 76.7% + 56.7% = 133.4%). If the collision rate is the fraction of episodes with at least one collision, this must be stated explicitly; if a trial can both succeed and collide, the current presentation makes it difficult to interpret the ablation's advantage. Please clarify the definition and, ideally, report a single outcome per trial (success, collision, timeout) or a confusion-style breakdown.
  4. [3.3 / Appendix A-B] Several parameters needed to reproduce PD-RiskNet and the reward are missing. In particular, the vertical angle threshold θ in Section 3.3.1, the FPS sample count, the distal downsampling factor, the distance cap d_max, the avoidance coefficient α_avoid, the velocity-tracking weight β_va, and the privileged-height supervision loss or weight are not specified. Nhist=10 appears only in Appendix B, and dthresh=1m and Nsec=36 appear in the main text. Please provide a complete hyperparameter table or list all numeric values in one place, otherwise the core architecture cannot be reimplemented from the paper alone.
minor comments (6)
  1. [1] In the contribution list on page 2, item 2 is incomplete: '2. Novel LiDAR Perception Network (PD-RiskNet):' is followed by no description. Please complete the sentence.
  2. [Figure 1 caption] The caption contains a typo, 'Omni-Perceptio', which should read 'Omni-Perception'.
  3. [4.1] The sentence 'You can find more lidar model scan patterns in the Appendix.??' contains a stray '??' placeholder and should be completed or removed.
  4. [4.4] Please clarify which robot platform(s) were used in the real-world experiments: Figure 2 shows the Unitree Go2, while Figure 4 shows the Unitree G1. The text should state the exact hardware used for each reported trial.
  5. [3.3.1 / B.1] The privileged-height supervision is mentioned as a training signal, but no loss function, prediction target, or loss weight is described. Please specify how this supervision is applied.
  6. [3.3.2] The reward function already contains a hand-designed LiDAR-based avoidance velocity, while the paper claims that the policy learns directly from raw point clouds. This is not circular, but the relationship should be stated explicitly: the observation is raw, yet the reward injects a LiDAR-derived prior, so the learned perception is shaped by that prior. Please add a clarifying sentence.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: real-world validation and independent evidence support the core claims.

full rationale

The paper's derivation chain is not circular. The policy is trained with rewards that use LiDAR sector distances (r_vel_avoid with V_avoid computed from sector minima) and ray-distance maximization (r_rays), but these are training objectives, not reported predictions. The paper's evidence for avoidance is real-world deployment on a physical Unitree robot with a physical Livox Mid-360 sensor (Table 4), which is not forced by construction from the simulator or the reward equations. The LiDAR simulator fidelity validation (Sec. 4.1) is qualitative and could be stronger, but that is an evidence-strength concern, not circularity. The reward design does handcraft the avoidance direction (a classical potential-field-like term), yet the policy must still learn to track it through raw point-cloud features and transfer that behavior to a real robot; the success rate on real obstacles is an independent test. Self-citations appear only in passing related-work and manipulation contexts and are not load-bearing for the central claim. Finally, the Limitations section explicitly predicts that averaging distant point clouds will destroy features of very small objects, and Table 4 shows 83% success on small obstacles; this consistency is an honest acknowledgment of a design trade-off, not a circular step. No equation in the paper defines the claimed result in terms of the inputs, and no fitted parameter is renamed as a prediction.

Assumptions & free parameters 11 free parameters · 5 assumptions · 4 invented entities

The central claim depends on a set of hand-chosen reward parameters and architecture dimensions. The reward terms themselves are tunable free parameters. No invented physical entities are introduced; the new items are a perception network, a simulator, and a reward design, all testable. The weakest element is the absence of reported values for several important hyperparameters and no sensitivity analysis for the most critical ones (theta, dthresh, alpha_avoid, dmax).

free parameters (11)
  • Vertical angle threshold theta = Not specified
    Partition of point cloud into proximal and distal regions. No sensitivity analysis is reported.
  • FPS sample count for proximal cloud = Not specified
    Number of points kept after farthest point sampling is not stated in the available text, but it determines the input size of the proximal GRU.
  • Distal downsampling factor = Not specified
    Average downsampling factor for distal points is not reported.
  • Sector count Nsec=36 = 36
    Number of angular sectors for avoidance velocity computation in the reward. Chosen by hand.
  • Avoidance threshold dthresh=1m = 1.0 m
    Distance below which the avoidance velocity activates in the reward. Hand-chosen.
  • Avoidance exponential coefficient alpha_avoid = Not specified
    Even though the text says its magnitude is calculated based on proximity, the exact value of alpha_avoid is not given.
  • Velocity tracking weight beta_va = Not specified in the main text; weight for the reward is 2
    Must be tuned; the reward table gives the weight but not the internal beta_va value.
  • Distance cap dmax for ray reward = Not specified
    Capping distance for the distance maximization reward is not given.
  • Proximal GRU embedding dimension = 187
    Architectural choice, likely hand-tuned.
  • Distal GRU embedding dimension = 64
    Architectural choice, likely hand-tuned.
  • Nhist=10 frames = 10
    Observation history length, chosen by hand.
assumptions (5)
  • standard math Standard reinforcement learning assumptions: MDP with Markovian state, PPO with an infinite-horizon discounted return.
    Invoked in Section 3.1 and Appendix C. Standard RL framework with no special epistemological commitment beyond the standard assumptions needed for PPO.
  • domain assumption LiDAR point cloud input contains enough information for collision avoidance.
    The whole method assumes raw LiDAR measurements are sufficient for safe locomotion. The paper does not prove this; it demonstrates it empirically in the tested scenarios.
  • domain assumption The reward function encodes the desired behavior of avoiding obstacles and following commands.
    The velocity tracking with avoidance reward and the distance maximization reward are hand-designed and inserted into the MDP in Section 3.3.2. If they do not capture the true objective, the policy will not perform as claimed.
  • domain assumption Sim-to-real transfer works because the simulator models the LiDAR and physics closely enough.
    The sim-to-real claim depends on the fidelity of the custom LiDAR simulation and the physics domain randomization. The paper supports this only qualitatively in figure 4 and with zero-shot real-world results.
  • domain assumption The proprioceptive state and physics simulator correctly model the Unitree G1 robot dynamics.
    Used throughout training in Isaac Gym. The paper relies on the standard Unitree G1 model in the simulator without a dedicated model verification section.
invented entities (4)
  • PD-RiskNet (Proximal-Distal Risk-Aware Hierarchical Network) independent evidence
    purpose: Perception module that processes raw spatio-temporal LiDAR point clouds into features for the locomotion policy.
    It is a new architecture, but its effect is measured in the paper: the ablation shows it outperforms simpler alternatives. That is evidence, and the architecture itself can be re-used.
  • New LiDAR simulation toolkit with fast raycasting and realistic noise independent evidence
    purpose: Train policies in parallel with realistic LiDAR, including non-repetitive scans and self-occlusion.
    The toolkit is claimed to be released on GitHub and is compared against Isaac Sim in Table 1. If the code is public, it is directly reusable.
  • Avoidance velocity V_avoid as a reward term independent evidence
    purpose: Encourages the robot to move away from nearby obstacles by combining sector-based repulsive velocity with the user command.
    This is a new reward design that can be tested independently by other groups, and is not a hidden physical entity.
  • Privileged height supervision signal independent evidence
    purpose: Supervises the proximal GRU during training to learn terrain-relevant features.
    Standard technique in legged locomotion; its effect is measured in the ablation only indirectly through the overall success rate.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Omni-Perception: Omnidirectional Collision Avoidance for Legged Locomotion in Dynamic Environments." pith.science (2026). https://pith.science/paper/X5ES7BET

@misc{pith2026250519214,
  author       = {Pith},
  title        = {Pith review of: Omni-Perception: Omnidirectional Collision Avoidance for Legged Locomotion in Dynamic Environments},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/X5ES7BET}},
  note         = {Machine review of arXiv:2505.19214}
}
read the original abstract

Agile locomotion in complex 3D environments requires robust spatial awareness to safely avoid diverse obstacles such as aerial clutter, uneven terrain, and dynamic agents. Depth-based perception approaches often struggle with sensor noise, lighting variability, computational overhead from intermediate representations (e.g., elevation maps), and difficulties with non-planar obstacles, limiting performance in unstructured environments. In contrast, direct integration of LiDAR sensing into end-to-end learning for legged locomotion remains underexplored. We propose Omni-Perception, an end-to-end locomotion policy that achieves 3D spatial awareness and omnidirectional collision avoidance by directly processing raw LiDAR point clouds. At its core is PD-RiskNet (Proximal-Distal Risk-Aware Hierarchical Network), a novel perception module that interprets spatio-temporal LiDAR data for environmental risk assessment. To facilitate efficient policy learning, we develop a high-fidelity LiDAR simulation toolkit with realistic noise modeling and fast raycasting, compatible with platforms such as Isaac Gym, Genesis, and MuJoCo, enabling scalable training and effective sim-to-real transfer. Learning reactive control policies directly from raw LiDAR data enables the robot to navigate complex environments with static and dynamic obstacles more robustly than approaches relying on intermediate maps or limited sensing. We validate Omni-Perception through real-world experiments and extensive simulation, demonstrating strong omnidirectional avoidance capabilities and superior locomotion performance in highly dynamic environments.

Figures

Figures reproduced from arXiv: 2505.19214 by the authors.

Figure 1
Figure 1. Validation scenarios for the Omni-Perception framework. Effective omnidirectional collision avoid￾ance is demonstrated on the left, where the robot reacts to obstacles from various approach vectors. Robustness against diverse environmental features is shown on the right, including successful negotiation of aerial, trans￾parent, slender, and ground obstacles. These results highlight the capacity of the Omni-Perceptio… view at source ↗
Figure 2
Figure 2. Proposed System Framework. (a) Visualization of differing sensor coverage: the typically narrow, forward-directed field of view of a depth camera (top) contrasted with the broader and longer range, coverage of a LiDAR sensor (bottom), shown on the Unitree Go2 robot. (b) Detailed diagram of the perception and control pipeline. Raw point cloud history is processed via two pathways within PD-RiskNet (average downsampli… view at source ↗
Figure 3
Figure 3. The calculation of the sector￾based avoidance velocity. The avoidance velocity V avoid t actively pushes the robot away from nearby obstacles detected by LiDAR. The robot’s 360° horizontal surroundings are divided into Nsec = 36 angular sectors. Within each sector j, the minimum obstacle distance d j t is found. If d j t is below a threshold d thresh (e.g., 1m), an avoidance velocity com￾ponent V avoid t,j is genera… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Comparison of simulated and real point cloud for the Unitree G1 robot. (a) The physical Unitree G1 robot setup. (b) Real-world LiDAR scan captured by the onboard Livox Mid-360 sensor. (c) (Ours) Point cloud generated using our Livox Mid-360 sensor model within the Isaa…
Figure 6
Figure 6. Figure 6: Simulation ablation experiment scene set￾ting. Scenario Omni-Perception Unitree System Static obstacles 30/30 (100%) 30/30 (100%) Aerial obstacles 21/30 (70%) 0/30 (0%) Small obstacles 25/30 (83%) 30/30 (100%) Moving humans 27/30 (90%) 0/30 (0%) [PITH_FULL_IMAGE:figur…
Figure 5
Figure 5. Figure 5: Robot obstacle avoidance performance was assessed in varied scenarios, including complex terrain and dynamic human interference. Goal [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 7
Figure 7. Figure 7: dense grass. Because the entire passage was narrow, the robot was forced to enter the grass on the right side of the picture. After entering the grass, the robot’s surroundings were perceived as a dangerous area, which led to mission failure. Objects that are too small…
Figure 8
Figure 8. Figure 8: Thin rod. Since we will average the distant point clouds, the fea￾tures of very small objects will be destroyed, making it impossible to respond correctly to such obstacles. 9 [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 4 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. CBF-RL: Safety Filtering Reinforcement Learning in Training with Control Barrier Functions

    cs.RO 2025-10 conditional novelty 6.0 of 10

    Training RL policies with a closed-form CBF safety filter plus CBF reward lets a Unitree G1 humanoid avoid obstacles and climb stairs without a runtime safety filter.

  2. MM-Nav: Multi-View VLA Model for Robust Visual Navigation via Multi-Expert Learning

    cs.RO 2025-10 conditional novelty 6.0 of 10

    A four-camera VLA navigation model trained by distilling multiple RL experts achieves strong simulation performance and qualitative real-world transfer.

  3. End-to-End Humanoid Robot Safe and Comfortable Locomotion Policy

    cs.RO 2025-08 conditional novelty 5.0 of 10

    An end-to-end humanoid locomotion policy maps raw LiDAR point clouds to motor commands using P3O with CBF-inspired safety costs and comfort rewards, with sim-to-real tests on a Unitree G1.

  4. Humanoid Occupancy: Enabling A Generalized Multimodal Occupancy Perception System on Humanoid Robots

    cs.RO 2025-07 conditional novelty 5.0 of 10

    A humanoid-specific multimodal occupancy perception system with a new dataset, sensor layout, and a fusion network that claims state-of-the-art results on its own benchmark.

Reference graph

Works this paper leans on

63 extracted references · 42 canonical work pages · cited by 4 Pith papers

  1. [1]

    A. Elfes. Using occupancy grids for mobile robot perception and navigation. Computer, 22 (6):46–57, 1989

    work page 1989

  2. [2]

    J. Lee, J. Hwangbo, L. Wellhausen, V . Koltun, and M. Hutter. Learning quadrupedal locomo- tion over challenging terrain. Science robotics, 5(47):eabc5986, 2020

    work page 2020

  3. [3]

    T. He, C. Zhang, W. Xiao, G. He, C. Liu, and G. Shi. Agile but safe: Learning collision-free high-speed legged locomotion. In Robotics: Science and Systems (RSS) , 2024

    work page 2024

  4. [4]

    Z. Xu, B. Chen, X. Zhan, Y . Xiu, C. Suzuki, and K. Shimada. A vision-based autonomous uav inspection framework for unknown tunnel construction sites with dynamic obstacles. IEEE Robotics and Automation Letters, 8(8):4983–4990, 2023

    work page 2023

  5. [5]

    D. Kim, D. Carballo, J. Di Carlo, B. Katz, G. Bledt, B. Lim, and S. Kim. Vision aided dy- namic exploration of unstructured terrain with a small-scale quadruped robot. In 2020 IEEE International Conference on Robotics and Automation (ICRA) , pages 2464–2470, 2020. doi: 10.1109/ICRA40945.2020.9196777

    arXiv 2020

  6. [6]

    Dudzik, M

    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 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages 3664–3671. IEEE, 2020

    work page 2020

  7. [7]

    G. B. Margolis and P. Agrawal. Walk these ways: Tuning robot control for generalization with multiplicity of behavior. Conference on Robot Learning, 2022

    work page 2022

  8. [8]

    Margolis, G

    G. Margolis, G. Yang, K. Paigwar, T. Chen, and P. Agrawal. Rapid locomotion via reinforce- ment learning. In Robotics: Science and Systems , 2022

    work page 2022

Show all 63 references
  1. [9]

    Hwangbo, J

    J. Hwangbo, J. Lee, A. Dosovitskiy, D. Bellicoso, V . Tsounis, V . Koltun, and M. Hutter. Learn- ing agile and dynamic motor skills for legged robots. sci. Robotics, 4:26, 2019

    2019

  2. [10]

    Rudin, D

    N. Rudin, D. Hoeller, P. Reist, and M. Hutter. Learning to walk in minutes using massively parallel deep reinforcement learning. InConference on Robot Learning, pages 91–100. PMLR, 2022

    2022

  3. [11]

    Kumar, Z

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

    2021

  4. [12]

    Z. Wang, Y . Jia, L. Shi, H. Wang, H. Zhao, X. Li, J. Zhou, J. Ma, and G. Zhou. Arm-constrained curriculum learning for loco-manipulation of a wheel-legged robot. In 2024 IEEE/RSJ Inter- national Conference on Intelligent Robots and Systems (IROS) , pages 10770–10776. IEEE, 2024

    2024

  5. [13]

    Cheng, K

    X. Cheng, K. Shi, A. Agarwal, and D. Pathak. Extreme parkour with legged robots. In 2024 IEEE International Conference on Robotics and Automation (ICRA) , pages 11443–11450. IEEE, 2024

    2024

  6. [14]

    Kareer, N

    S. Kareer, N. Yokoyama, D. Batra, S. Ha, and J. Truong. Vinl: Visual navigation and loco- motion over obstacles. In 2023 IEEE International Conference on Robotics and Automation (ICRA), pages 2018–2024. IEEE, 2023. 10

    2023

  7. [15]

    Agarwal, A

    A. Agarwal, A. Kumar, J. Malik, and D. Pathak. Legged locomotion in challenging terrains using egocentric vision. In Conference on robot learning, pages 403–415. PMLR, 2023

    2023

  8. [16]

    Plagemann, S

    C. Plagemann, S. Mischke, S. Prentice, K. Kersting, N. Roy, and W. Burgard. Learning predic- tive terrain models for legged robot locomotion. In 2008 IEEE/RSJ International Conference on Intelligent Robots and Systems , pages 3545–3552. IEEE, 2008

    2008

  9. [17]

    Cheng, Y

    A.-C. Cheng, Y . Ji, Z. Yang, Z. Gongye, X. Zou, J. Kautz, E. Bıyık, H. Yin, S. Liu, and X. Wang. Navila: Legged robot vision-language-action model for navigation. arXiv preprint arXiv:2412.04453, 2024

    2024 arXiv

  10. [18]

    T. Miki, L. Wellhausen, R. Grandia, F. Jenelten, T. Homberger, and M. Hutter. Elevation map- ping for locomotion and navigation using gpu. In 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages 2273–2280. IEEE, 2022

    2022

  11. [19]

    Li and J

    Y . Li and J. Ibanez-Guzman. Lidar for autonomous driving: The principles, challenges, and trends for automotive lidar and perception systems. IEEE Signal Processing Magazine, 37(4): 50–61, 2020

    2020

  12. [20]

    Y . Li, L. Ma, Z. Zhong, F. Liu, M. A. Chapman, D. Cao, and J. Li. Deep learning for lidar point clouds in autonomous driving: A review. IEEE Transactions on Neural Networks and Learning Systems, 32(8):3412–3432, 2020

    2020

  13. [21]

    T. Ma, Z. Wang, J. Zhou, M. Wang, and J. Liang. Glover: Generalizable open-vocabulary affordance reasoning for task-oriented grasping. arXiv preprint arXiv:2411.12286, 2024

    2024 arXiv

  14. [22]

    T. Ma, J. Zheng, Z. Wang, Z. Gao, J. Zhou, and J. Liang. Glover++: Unleashing the poten- tial of affordance learning from human behaviors for robotic manipulation. arXiv preprint arXiv:2505.11865, 2025

    2025 arXiv

  15. [23]

    T. Ma, J. Zhou, Z. Wang, R. Qiu, and J. Liang. Contrastive imitation learning for language- guided multi-task robotic manipulation. arXiv preprint arXiv:2406.09738, 2024

    2024 arXiv

  16. [24]

    J. Zhou, K. Ye, J. Liu, T. Ma, Z. Wang, R. Qiu, K.-Y . Lin, Z. Zhao, and J. Liang. Exploring the limits of vision-language-action manipulations in cross-task generalization. arXiv preprint arXiv:2505.15660, 2025

    2025

  17. [25]

    J. Zhou, T. Ma, K.-Y . Lin, Z. Wang, R. Qiu, and J. Liang. Mitigating the human-robot domain discrepancy in visual pre-training for robotic manipulation. arXiv preprint arXiv:2406.14235, 2024

    2024 arXiv

  18. [26]

    Y . Jia, G. Wang, Y . Dong, J. Wu, Y . Zeng, H. Lin, Z. Wang, H. Ge, W. Gu, K. Ding, et al. Discoverse: Efficient robot simulation in complex high-fidelity environments. arXiv preprint arXiv:2507.21981, 2025

    2025 arXiv

  19. [27]

    C. D. Bellicoso, M. Bjelonic, L. Wellhausen, K. Holtmann, F. G ¨unther, M. Tranzatto, P. Fankhauser, and M. Hutter. Advances in real-world applications for legged robots. Journal of Field Robotics, 35(8):1311–1326, 2018

    2018

  20. [28]

    Delmerico, S

    J. Delmerico, S. Mintchev, A. Giusti, B. Gromov, K. Melo, T. Horvat, C. Cadena, M. Hutter, A. Ijspeert, D. Floreano, et al. The current state and future outlook of rescue robotics. Journal of Field Robotics, 36(7):1171–1191, 2019

    2019

  21. [29]

    Rudin, D

    N. Rudin, D. Hoeller, M. Bjelonic, and M. Hutter. Advanced skills by learning locomotion and local navigation end-to-end. In 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 2497–2503. IEEE, 2022

    2022

  22. [30]

    Wisth, M

    D. Wisth, M. Camurri, and M. Fallon. Vilens: Visual, inertial, lidar, and leg odometry for all-terrain legged robots. IEEE Transactions on Robotics, 39(1):309–326, 2022. 11

    2022

  23. [31]

    Y . J. Ma, W. Liang, G. Wang, D.-A. Huang, O. Bastani, D. Jayaraman, Y . Zhu, L. Fan, and A. Anandkumar. Eureka: Human-level reward design via coding large language models.arXiv preprint arXiv:2310.12931, 2023

    2023 arXiv

  24. [32]

    Zucker, N

    M. Zucker, N. Ratliff, M. Stolle, J. Chestnutt, J. A. Bagnell, C. G. Atkeson, and J. Kuffner. Optimization and learning for rough terrain legged locomotion. The International Journal of Robotics Research, 30(2):175–191, 2011

    2011

  25. [33]

    S. Choi, G. Ji, J. Park, H. Kim, J. Mun, J. H. Lee, and J. Hwangbo. Learning quadrupedal locomotion on deformable terrain. Science Robotics, 8(74):eade2256, 2023

    2023

  26. [34]

    Hoeller, N

    D. Hoeller, N. Rudin, D. Sako, and M. Hutter. Anymal parkour: Learning agile navigation for quadrupedal robots. Science Robotics, 9(88):eadi7566, 2024

    2024

  27. [35]

    J. Lee, M. Bjelonic, A. Reske, L. Wellhausen, T. Miki, and M. Hutter. Learning robust au- tonomous navigation and locomotion for wheeled-legged robots. Science Robotics , 9(89): eadi9641, 2024

    2024

  28. [36]

    Zhuang, Z

    Z. Zhuang, Z. Fu, J. Wang, C. Atkeson, S. Schwertfeger, C. Finn, and H. Zhao. Robot parkour learning. arXiv preprint arXiv:2309.05665, 2023

    2023 arXiv

  29. [37]

    Zhuang, S

    Z. Zhuang, S. Yao, and H. Zhao. Humanoid parkour learning. arXiv preprint arXiv:2406.10759, 2024

    2024 arXiv

  30. [38]

    Chane-Sane, J

    E. Chane-Sane, J. Amigo, T. Flayols, L. Righetti, and N. Mansard. Soloparkour: Constrained reinforcement learning for visual locomotion from privileged experience. In Conference on Robot Learning (CoRL), 2024

    2024

  31. [39]

    T. Miki, J. Lee, J. Hwangbo, L. Wellhausen, V . Koltun, and M. Hutter. Learning robust percep- tive locomotion for quadrupedal robots in the wild. Science robotics, 7(62):eabk2822, 2022

    2022

  32. [40]

    Hoeller, L

    D. Hoeller, L. Wellhausen, F. Farshidian, and M. Hutter. Learning a state representation and navigation in cluttered and dynamic environments. IEEE Robotics and Automation Letters , 6 (3):5081–5088, 2021

    2021

  33. [41]

    Y . F. Chen, M. Liu, M. Everett, and J. P. How. Decentralized non-communicating multiagent collision avoidance with deep reinforcement learning. In 2017 IEEE international conference on robotics and automation (ICRA) , pages 285–292. IEEE, 2017

    2017

  34. [42]

    P. Arm, G. Waibel, J. Preisig, T. Tuna, R. Zhou, V . Bickel, G. Ligeza, T. Miki, F. Kehl, H. Kol- venbach, et al. Scientific exploration of challenging planetary analog environments with a team of legged robots. Science robotics, 8(80):eade9548, 2023

    2023

  35. [43]

    Wermelinger, P

    M. Wermelinger, P. Fankhauser, R. Diethelm, P. Kr¨usi, R. Siegwart, and M. Hutter. Navigation planning for legged robots in challenging terrain. In 2016 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages 1184–1189. IEEE, 2016

    2016

  36. [44]

    H. Yin, X. Xu, S. Lu, X. Chen, R. Xiong, S. Shen, C. Stachniss, and Y . Wang. A survey on global lidar localization: Challenges, advances and open problems. International Journal of Computer Vision, 132(8):3139–3171, 2024

    2024

  37. [45]

    Haddadin, A

    S. Haddadin, A. De Luca, and A. Albu-Sch ¨affer. Robot collisions: A survey on detection, isolation, and identification. IEEE Transactions on Robotics, 33(6):1292–1312, 2017

    2017

  38. [46]

    Gaertner, M

    M. Gaertner, M. Bjelonic, F. Farshidian, and M. Hutter. Collision-free mpc for legged robots in static and dynamic scenes. In 2021 IEEE International Conference on Robotics and Au- tomation (ICRA), pages 8266–8272. IEEE, 2021

    2021

  39. [47]

    Park, H.-D

    J.-W. Park, H.-D. Oh, and M.-J. Tahk. Uav collision avoidance based on geometric approach. In 2008 SICE Annual Conference, pages 2122–2126. IEEE, 2008. 12

    2008

  40. [48]

    Lin and S

    M. Lin and S. Gottschalk. Collision detection between geometric models: A survey. In Proc. of IMA conference on mathematics of surfaces , volume 1, pages 602–608, 1998

    1998

  41. [49]

    Chiu, J.-P

    J.-R. Chiu, J.-P. Sleiman, M. Mittal, F. Farshidian, and M. Hutter. A collision-free mpc for whole-body dynamic locomotion and manipulation. In 2022 international conference on robotics and automation (ICRA), pages 4686–4693. IEEE, 2022

    2022

  42. [50]

    Lindqvist, S

    B. Lindqvist, S. S. Mansouri, A.-a. Agha-mohammadi, and G. Nikolakopoulos. Nonlinear mpc for collision avoidance and control of uavs with dynamic obstacles. IEEE robotics and automation letters, 5(4):6001–6008, 2020

    2020

  43. [51]

    Q. Liao, Z. Li, A. Thirugnanam, J. Zeng, and K. Sreenath. Walking in narrow spaces: Safety- critical locomotion control for quadrupedal robots with duality-based optimization, 2023. URL https://arxiv.org/abs/2212.14199

    2023 arXiv

  44. [52]

    Koptev, N

    M. Koptev, N. Figueroa, and A. Billard. Reactive collision-free motion generation in joint space via dynamical systems and sampling-based mpc. The International Journal of Robotics Research, 43(13):2049–2069, 2024

    2024

  45. [53]

    Z. Xu, X. Han, H. Shen, H. Jin, and K. Shimada. Navrl: Learning safe flight in dynamic environments. IEEE Robotics and Automation Letters , 10(4):3668–3675, 2025. doi:10.1109/ LRA.2025.3546069

    2025

  46. [54]

    Y . Song, K. Shi, R. Penicka, and D. Scaramuzza. Learning perception-aware agile flight in cluttered environments, 2022. URL https://arxiv.org/abs/2210.01841

    2022 arXiv

  47. [55]

    Makoviychuk, L

    V . Makoviychuk, L. Wawrzyniak, Y . Guo, M. Lu, K. Storey, M. Macklin, D. Hoeller, N. Rudin, A. Allshire, A. Handa, and G. State. Isaac gym: High performance gpu-based physics simula- tion for robot learning, 2021

    2021

  48. [56]

    Koenig and A

    N. Koenig and A. Howard. Design and use paradigms for gazebo, an open-source multi-robot simulator. In 2004 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) (IEEE Cat. No.04CH37566) , volume 3, pages 2149–2154 vol.3, 2004. doi:10.1109/ IROS.2004.1389727

    2004 arXiv

  49. [57]

    G. Authors. Genesis: A universal and generative physics engine for robotics and beyond, December 2024. URL https://github.com/Genesis-Embodied-AI/Genesis

    2024

  50. [58]

    Kulkarni, W

    M. Kulkarni, W. Rehberg, and K. Alexis. Aerial gym simulator: A framework for highly parallelized simulation of aerial robots. IEEE Robotics and Automation Letters , 2025

    2025

  51. [59]

    M. Macklin. Warp: A high-performance python framework for gpu simulation and graphics. https://github.com/nvidia/warp, March 2022. NVIDIA GPU Technology Conference (GTC)

    2022

  52. [60]

    Hu, T.-M

    Y . Hu, T.-M. Li, L. Anderson, J. Ragan-Kelley, and F. Durand. Taichi: a language for high- performance computation on spatially sparse data structures. ACM Transactions on Graphics (TOG), 38(6):201, 2019

    2019

  53. [61]

    Lauterbach, M

    C. Lauterbach, M. Garland, S. Sengupta, D. Luebke, and D. Manocha. Fast bvh construction on gpus. In Computer Graphics F orum, volume 28, pages 375–384. Wiley Online Library, 2009

    2009

  54. [62]

    C. R. Qi, L. Yi, H. Su, and L. J. Guibas. Pointnet++: Deep hierarchical feature learning on point sets in a metric space. arXiv preprint arXiv:1706.02413, 2017

    2017 arXiv

  55. [63]

    Schulman, F

    J. Schulman, F. Wolski, P. Dhariwal, A. Radford, and O. Klimov. Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347, 2017. 13 A Implementation Details A.1 Rewards Term Equation Weight Omni-Perception Rewards Velocity Tracking with Avoidance (rvel,avoid ) e...

    2017 arXiv

Pith tools

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