pith. sign in

arxiv: 2606.23153 · v1 · pith:OM3UZ4LRnew · submitted 2026-06-22 · 💻 cs.RO

Asymmetric physics enables efficient learning in quadrupedal robot swarms

Yuang Zhang , Yunlong Song , Zhihao He , Zelin Ni , Kangyu Wang , Tianchi Liu , Yu Hu , Feng Yu
show 2 more authors
Danping Zou Weiyao Lin
This is my paper

Pith reviewed 2026-06-26 08:29 UTC · model grok-4.3

classification 💻 cs.RO
keywords asymmetric physicsquadrupedal robot swarmsdecentralized controlend-to-end learningvision-based navigationdifferentiable surrogate modelsmulti-robot coordinationcontact-rich locomotion
0
0 comments X

The pith

Asymmetric physics allows efficient end-to-end learning of decentralized vision-based control for large quadrupedal robot swarms.

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

The paper shows that asymmetric physics—pairing a high-fidelity non-differentiable simulator for realistic robot motion and contacts with differentiable surrogate models for policy gradients—makes it possible to train vision-based decentralized controllers for swarms of up to 512 quadrupeds. This separation addresses the inefficiency of standard reinforcement learning when visual perception, dense robot-robot interactions, and contact-rich locomotion must be learned together. The resulting policies enable coordinated navigation in cluttered environments without explicit communication or central planning. The policies generalize across multiple environment types and transfer zero-shot to physical robots in real-world scenarios.

Core claim

Asymmetric physics enables efficient end-to-end learning of vision-based, decentralized control in large swarms of quadrupedal robots. During training, quadrupeds interact in shared environments, where a high-fidelity, non-differentiable simulator generates realistic motion and contact dynamics, and differentiable surrogate models provide gradients for navigation and locomotion policies. This separation enables up to 512 quadrupeds to learn coordinated navigation policies in obstacle-rich environments. At deployment, each robot acts from a single forward-facing depth camera, without explicit communication, centralized planning, or global maps. The policies generalize across forests, bridges,

What carries the argument

asymmetric physics, the separation of a high-fidelity non-differentiable simulator for generating realistic motion and contact dynamics from differentiable surrogate models that supply gradients for the navigation and locomotion policies.

Load-bearing premise

The differentiable surrogate models must supply gradients that remain sufficiently accurate and unbiased relative to the high-fidelity non-differentiable simulator across the full range of contact-rich interactions and visual observations encountered during training.

What would settle it

Training runs that compare symmetric physics setups (fully non-differentiable or fully differentiable) against the asymmetric setup and show that only the asymmetric version scales policies successfully to 512 robots with successful zero-shot real-world transfer would support the claim; failure of the asymmetric method to produce scalable or transferable policies would falsify it.

read the original abstract

Animal collectives navigate cluttered environments through local coordination, yet robot swarms still struggle to reproduce this capability in the physical world. End-to-end learning offers a route to such coordination, but scaling it to embodied swarms remains difficult: standard sampling-based reinforcement learning becomes inefficient when visual perception, dense robot-robot interaction, and contact-rich locomotion must be learned together. Here we show that asymmetric physics enables efficient end-to-end learning of vision-based, decentralized control in large swarms of quadrupedal robots. During training, quadrupeds interact in shared environments, where a high-fidelity, non-differentiable simulator generates realistic motion and contact dynamics, and differentiable surrogate models provide gradients for navigation and locomotion policies. This separation enables up to 512 quadrupeds to learn coordinated navigation policies in obstacle-rich environments. At deployment, each robot acts from a single forward-facing depth camera, without explicit communication, centralized planning, or global maps. The policies generalize across forests, bridges, enclosures, narrow passages, and mazes, and zero-shot transfer to six physical quadrupeds across five real-world scenarios. The resulting swarms exhibit predictive avoidance, right-side yielding, pausing before bottlenecks, and wall following, showing that asymmetric physics enables efficient training of scalable decentralized control policies for quadrupedal robot swarms.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript claims that an asymmetric physics setup—pairing a high-fidelity non-differentiable simulator for motion and contact dynamics with differentiable surrogate models for policy gradients—enables efficient end-to-end learning of vision-based decentralized control for swarms of up to 512 quadrupedal robots. The resulting policies generalize across multiple simulated environments and achieve zero-shot transfer to six physical robots in five real-world scenarios, producing emergent behaviors such as predictive avoidance and right-side yielding without explicit communication or centralized planning.

Significance. If the central empirical claim is supported by rigorous validation, the result would be significant for scalable embodied swarm learning: it offers a concrete route to training contact-rich, vision-based multi-agent policies that standard sampling-based RL cannot handle efficiently at this scale. The reported zero-shot sim-to-real transfer and emergence of coordinated local behaviors would constitute a notable demonstration in the field.

major comments (2)
  1. [Methods / Training Setup] The description of the differentiable surrogate models supplies no quantitative error metrics, bias analysis, or ablation studies on gradient fidelity relative to the high-fidelity simulator for contact-rich regimes (multi-robot collisions, terrain contacts, depth observations). This assumption is load-bearing for the scalability claim to 512 agents and the reported emergence of yielding behaviors.
  2. [Results] The results section reports generalization across forests, bridges, enclosures, narrow passages, and mazes together with zero-shot transfer to physical robots, yet provides no quantitative performance metrics, success rates, or statistical comparisons against baselines that would allow assessment of the strength of these claims.
minor comments (2)
  1. Notation for the surrogate models and the precise interface between the non-differentiable simulator and the gradient-providing surrogates should be defined more explicitly, ideally with a diagram or pseudocode.
  2. The abstract states specific numbers (512 quadrupeds, six physical robots, five scenarios); the main text should include corresponding tables or figures with error bars or confidence intervals.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which highlight important gaps in validation and quantification. We address each major point below and commit to revisions that strengthen the manuscript without altering its core claims.

read point-by-point responses

  1. Referee: [Methods / Training Setup] The description of the differentiable surrogate models supplies no quantitative error metrics, bias analysis, or ablation studies on gradient fidelity relative to the high-fidelity simulator for contact-rich regimes (multi-robot collisions, terrain contacts, depth observations). This assumption is load-bearing for the scalability claim to 512 agents and the reported emergence of yielding behaviors.

    Authors: We agree that the manuscript currently lacks explicit quantitative validation of surrogate gradient fidelity in contact-rich regimes. In revision we will add error metrics (MSE between surrogate and finite-difference gradients on contact forces and depth), bias analysis, and ablations showing policy performance sensitivity to gradient accuracy. These will appear in a dedicated Methods subsection with supporting figures. revision: yes

  2. Referee: [Results] The results section reports generalization across forests, bridges, enclosures, narrow passages, and mazes together with zero-shot transfer to physical robots, yet provides no quantitative performance metrics, success rates, or statistical comparisons against baselines that would allow assessment of the strength of these claims.

    Authors: The referee is correct that aggregated quantitative metrics and baseline comparisons are absent. We will revise the Results section to report success rates, collision statistics, and navigation efficiency across environments, together with statistical comparisons (multiple seeds) against independent RL and non-surrogate baselines. Real-world transfer will include success counts over the five scenarios. revision: yes

Circularity Check

0 steps flagged

No circularity; empirical demonstration of asymmetric simulator setup

full rationale

The paper presents an empirical training method that separates a high-fidelity non-differentiable simulator (for motion and contacts) from differentiable surrogate models (for policy gradients). Reported outcomes consist of policy generalization across simulated environments and zero-shot transfer to physical robots. No equations, fitted parameters renamed as predictions, or self-citation chains are shown that reduce the central claim to its own inputs by construction. The result is a standard empirical robotics demonstration whose validity rests on external benchmarks (sim-to-real transfer) rather than internal definitional equivalence.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard assumptions of reinforcement learning and sim-to-real transfer plus the domain assumption that local depth images suffice for coordination without communication.

axioms (2)
  • domain assumption Standard RL assumptions (Markov property, reward shaping, policy gradient validity) hold for the multi-agent vision-based setting.
    Invoked implicitly by the use of end-to-end learning for navigation and locomotion policies.
  • domain assumption Local forward-facing depth observations contain sufficient information for decentralized coordination in the tested environments.
    Stated in the deployment description: each robot acts from a single forward-facing depth camera without explicit communication.

pith-pipeline@v0.9.1-grok · 5781 in / 1378 out tokens · 17148 ms · 2026-06-26T08:29:03.661458+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

64 extracted references · 3 canonical work pages

  1. [1]

    Adaptive behavior2(2), 189–218 (1993)

    Kube, C.R., Zhang, H.: Collective robotics: From social insects to robots. Adaptive behavior2(2), 189–218 (1993)

    1993

  2. [2]

    In: International Workshop on Swarm Robotics, pp

    Beni, G.: From swarm intelligence to swarm robotics. In: International Workshop on Swarm Robotics, pp. 1–9 (2004). Springer

    2004

  3. [3]

    In: International Workshop on Swarm Robotics, pp

    Şahin, E.: Swarm robotics: From sources of inspiration to domains of application. In: International Workshop on Swarm Robotics, pp. 10–20 (2004). Springer

    2004

  4. [4]

    Technical report (2019)

    Kang, C.-k., Fahimi, F., Griffin, R., Landrum, D.B., Mesmer, B., Zhang, G., Lee, T., Aono, H., Pohly, J., McCain, J., et al.: Marsbee-swarm of flapping wing flyers for enhanced mars exploration: Nasa innovative advanced concepts (niac)-phase i. Technical report (2019)

    2019

  5. [5]

    Science robotics8(80), 9548 (2023)

    Arm, P., Waibel, G., Preisig, J., Tuna, T., Zhou, R., Bickel, V., Ligeza, G., Miki, T., Kehl, F., Kolvenbach, H.,et al.: Scientific exploration of challenging planetary analog environments with a team of legged robots. Science robotics8(80), 9548 (2023)

    2023

  6. [6]

    Current opinion in biotechnology45, 76–84 (2017)

    Bayat, B., Crasta, N., Crespi, A., Pascoal, A.M., Ijspeert, A.: Environmental monitoring using autonomous vehicles: a survey of recent searching techniques. Current opinion in biotechnology45, 76–84 (2017)

    2017

  7. [7]

    IEEE Robotics & Automation Magazine19(1), 24–39 (2012)

    Dunbabin, M., Marques, L.: Robots for environmental monitoring: Significant advancements and applications. IEEE Robotics & Automation Magazine19(1), 24–39 (2012)

    2012

  8. [8]

    Autonomous Robots47(1), 77–93 (2023)

    Horyna, J., Baca, T., Walter, V., Albani, D., Hert, D., Ferrante, E., Saska, M.: Decentralized swarms of unmanned aerial vehicles for search and rescue operations without explicit communication. Autonomous Robots47(1), 77–93 (2023)

    2023

  9. [9]

    Science robotics3(14), 7650 (2018)

    Yang, G.-Z., Bellingham, J., Dupont, P.E., Fischer, P., Floridi, L., Full, R., Jacob- stein, N., Kumar, V., McNutt, M., Merrifield, R.,et al.: The grand challenges of science robotics. Science robotics3(14), 7650 (2018)

    2018

  10. [10]

    Science robotics5(49), 4385 (2020)

    Dorigo, M., Theraulaz, G., Trianni, V.: Reflections on the future of swarm robotics. Science robotics5(49), 4385 (2020)

    2020

  11. [11]

    In: Proceedings of the 14th Annual Conference on Computer Graphics and Interactive Techniques, pp

    Reynolds, C.W.: Flocks, herds and schools: A distributed behavioral model. In: Proceedings of the 14th Annual Conference on Computer Graphics and Interactive Techniques, pp. 25–34 (1987)

    1987

  12. [12]

    Physical review letters75(6), 1226 29 (1995)

    Vicsek, T., Czirók, A., Ben-Jacob, E., Cohen, I., Shochet, O.: Novel type of phase transition in a system of self-driven particles. Physical review letters75(6), 1226 29 (1995)

    1995

  13. [13]

    Nature433(7025), 513–516 (2005)

    Couzin, I.D., Krause, J., Franks, N.R., Levin, S.A.: Effective leadership and decision-making in animal groups on the move. Nature433(7025), 513–516 (2005)

    2005

  14. [14]

    In: Proceedings of the 9th Conference on Autonomous Robot Systems and Competitions, vol

    Mondada, F., Bonani, M., Raemy, X., Pugh, J., Cianci, C., Klaptocz, A., Mag- nenat, S., Zufferey, J.-C., Floreano, D., Martinoli, A.,et al.: The e-puck, a robot designed for education in engineering. In: Proceedings of the 9th Conference on Autonomous Robot Systems and Competitions, vol. 1, pp. 59–65 (2009). Castelo Branco: IPCB, Instituto Politécnico de ...

    2009

  15. [15]

    Science345(6198), 795–799 (2014)

    Rubenstein, M., Cornejo, A., Nagpal, R.: Programmable self-assembly in a thousand-robot swarm. Science345(6198), 795–799 (2014)

    2014

  16. [16]

    Foundations and Trends®in Robotics7(1-2), 1–179 (2018)

    Osa, T., Pajarinen, J., Neumann, G., Bagnell, J.A., Abbeel, P., Peters, J.,et al.: An algorithmic perspective on imitation learning. Foundations and Trends®in Robotics7(1-2), 1–179 (2018)

    2018

  17. [17]

    Adaptive computation and machine learning

    Sutton, R.S., Barto, A.G.: Reinforcement learning: an introduction, 2nd edn. Adaptive computation and machine learning. The MIT Press, Cambridge (2018)

    2018

  18. [18]

    Frontiers in Robotics and AI10, 1134841 (2023)

    Kuckling, J.: Recent trends in robot learning and evolution for swarm robotics. Frontiers in Robotics and AI10, 1134841 (2023)

    2023

  19. [19]

    Science Robotics4(26), 5872 (2019)

    Hwangbo, J., Lee, J., Dosovitskiy, A., Bellicoso, D., Tsounis, V., Koltun, V., Hutter, M.: Learning agile and dynamic motor skills for legged robots. Science Robotics4(26), 5872 (2019)

    2019

  20. [20]

    Science Robotics7(62), 2822 (2022)

    Miki, T., Lee, J., Hwangbo, J., Wellhausen, L., Koltun, V., Hutter, M.: Learn- ing robust perceptive locomotion for quadrupedal robots in the wild. Science Robotics7(62), 2822 (2022)

    2022

  21. [21]

    IEEE Robotics and Automation Letters6(3), 4257–4264 (2021)

    Fuchs, F., Song, Y., Kaufmann, E., Scaramuzza, D., Dürr, P.: Super-human per- formance in gran turismo sport using deep reinforcement learning. IEEE Robotics and Automation Letters6(3), 4257–4264 (2021)

    2021

  22. [22]

    Nature 602(7896), 223–228 (2022)

    Wurman, P.R., Barrett, S., Kawamoto, K., MacGlashan, J., Subramanian, K., Walsh, T.J., Capobianco, R., Devlic, A., Eckert, F., Fuchs, F.,et al.: Outrac- ing champion gran turismo drivers with deep reinforcement learning. Nature 602(7896), 223–228 (2022)

    2022

  23. [23]

    Nature 620(7976), 982–987 (2023)

    Kaufmann,E.,Bauersfeld,L.,Loquercio,A.,Müller,M.,Koltun,V.,Scaramuzza, D.: Champion-level drone racing using deep reinforcement learning. Nature 620(7976), 982–987 (2023)

    2023

  24. [24]

    Science Robotics8(82), 1462 (2023)

    Song, Y., Romero, A., Müller, M., Koltun, V., Scaramuzza, D.: Reaching the limit 30 in autonomous racing: Optimal control versus reinforcement learning. Science Robotics8(82), 1462 (2023)

    2023

  25. [25]

    Huh, D., Mohapatra, P.: Multi-agent Reinforcement Learning: A Comprehensive Survey. arXiv. arXiv:2312.10256 [cs] (2023). http://arxiv.org/abs/2312.10256 Accessed 2024-03-10

    arXiv 2023

  26. [26]

    Sensors23(7), 3625 (2023) https://doi.org/10.3390/ s23073625

    Orr, J., Dutta, A.: Multi-Agent Deep Reinforcement Learning for Multi-Robot Applications: A Survey. Sensors23(7), 3625 (2023) https://doi.org/10.3390/ s23073625 . Accessed 2024-10-08

    2023

  27. [27]

    Artificial Intelligence Review57(2), 41 (2024) https://doi.org/10

    Chung, J., Fayyad, J., Younes, Y.A., Najjaran, H.: Learning team-based nav- igation: a review of deep reinforcement learning techniques for multi-agent pathfinding. Artificial Intelligence Review57(2), 41 (2024) https://doi.org/10. 1007/s10462-023-10670-6 . Accessed 2024-10-08

    2024

  28. [28]

    Current Robotics Reports 3(4), 239–257 (2022) https://doi.org/10.1007/s43154-022-00091-8

    Wang, Y., Damani, M., Wang, P., Cao, Y., Sartoretti, G.: Distributed Rein- forcement Learning for Robot Teams: a Review. Current Robotics Reports 3(4), 239–257 (2022) https://doi.org/10.1007/s43154-022-00091-8 . Accessed 2024-03-14

    work page doi:10.1007/s43154-022-00091-8 2022

  29. [29]

    Basic Books, New York (1954)

    Piaget, J.: The Construction of Reality in the Child. Basic Books, New York (1954)

    1954

  30. [30]

    Perspectives on Psychological Science3(1), 2–13 (2008) https: //doi.org/10.1111/j.1745-6916.2008.00056.x

    Baillargeon, R.: Innate ideas revisited: For a principle of persistence in infants’ physical reasoning. Perspectives on Psychological Science3(1), 2–13 (2008) https: //doi.org/10.1111/j.1745-6916.2008.00056.x

    work page doi:10.1111/j.1745-6916.2008.00056.x 2008

  31. [31]

    arXiv preprint arXiv:2502.11831 (2025)

    Garrido,Q.,Ballas,N.,Assran,M.,Bardes,A.,Najman,L.,Rabbat,M.,Dupoux, E., LeCun, Y.: Intuitive physics understanding emerges from self-supervised pretraining on natural videos. arXiv preprint arXiv:2502.11831 (2025)

    arXiv 2025

  32. [32]

    Trends in cognitive sciences21(9), 649–665 (2017)

    Ullman, T.D., Spelke, E., Battaglia, P., Tenenbaum, J.B.: Mind games: Game engines as an architecture for intuitive physics. Trends in cognitive sciences21(9), 649–665 (2017)

    2017

  33. [33]

    20668–20696 (2022)

    Suh, H.J., Simchowitz, M., Zhang, K., Tedrake, R.: Do differentiable simulators give better policy gradients? In: International Conference on Machine Learning, pp. 20668–20696 (2022). PMLR

    2022

  34. [34]

    In: International Conference on Learning Representations (2021)

    Xu, J., Makoviychuk, V., Narang, Y., Ramos, F., Matusik, W., Garg, A., Mack- lin, M.: Accelerated policy learning with parallel differentiable simulation. In: International Conference on Learning Representations (2021)

    2021

  35. [35]

    Science Robotics3(20), 3536 (2018) 31

    Vásárhelyi, G., Virágh, C., Somorjai, G., Nepusz, T., Eiben, A.E., Vicsek, T.: Optimized flocking of autonomous drones in confined environments. Science Robotics3(20), 3536 (2018) 31

    2018

  36. [36]

    Science Robotics 7(66), 5954 (2022)

    Zhou, X., Wen, X., Wang, Z., Gao, Y., Li, H., Wang, Q., Yang, T., Lu, H., Cao, Y., Xu, C.,et al.: Swarm of micro flying robots in the wild. Science Robotics 7(66), 5954 (2022)

    2022

  37. [37]

    Springer, Berlin, Heidelberg (2008)

    Trianni, V.: Evolutionary Swarm Robotics: Evolving Self-organising Behaviours in Groups of Autonomous Robots. Springer, Berlin, Heidelberg (2008)

    2008

  38. [38]

    Swarm Intelligence8(2), 89–112 (2014)

    Francesca, G., Brambilla, M., Brutschy, A., Trianni, V., Birattari, M.: Automode: A novel approach to the automatic design of control software for robot swarms. Swarm Intelligence8(2), 89–112 (2014)

    2014

  39. [39]

    SN Computer Science3(2), 136 (2022)

    Kuckling, J., Van Pelt, V., Birattari, M.: Automode-cedrata: automatic design of behavior trees for controlling a swarm of robots with communication capabilities. SN Computer Science3(2), 136 (2022)

    2022

  40. [40]

    In: Exper- iments with the Mini-Robot Khepera, Proceedings of the First International Khepera Workshop, vol

    Mondada, F., Franzi, E., Guignard, A.: The development of khepera. In: Exper- iments with the Mini-Robot Khepera, Proceedings of the First International Khepera Workshop, vol. 1, pp. 7–14 (1999). sn

    1999

  41. [41]

    The International Journal of Robotics Research33(8), 1145–1161 (2014)

    Gauci, M., Chen, J., Li, W., Dodd, T.J., Groß, R.: Self-organized aggregation without computation. The International Journal of Robotics Research33(8), 1145–1161 (2014)

    2014

  42. [42]

    In: European Conference on Artificial Life, pp

    Garnier, S., Jost, C., Jeanson, R., Gautrais, J., Asadpour, M., Caprari, G., Ther- aulaz, G.: Aggregation behaviour as a source of collective decision in a group of cockroach-like-robots. In: European Conference on Artificial Life, pp. 169–178 (2005). Springer

    2005

  43. [43]

    PloS one11(3), 0151834 (2016)

    Duarte, M., Costa, V., Gomes, J., Rodrigues, T., Silva, F., Oliveira, S.M., Chris- tensen, A.L.: Evolution of collective behaviors for a real swarm of aquatic surface robots. PloS one11(3), 0151834 (2016)

    2016

  44. [44]

    In: OCEANS 2018 MTS/IEEE Charleston, pp

    Vallegra, F., Mateo, D., Tokić, G., Bouffanais, R., Yue, D.K.: Gradual collective upgrade of a swarm of autonomous buoys for dynamic ocean monitoring. In: OCEANS 2018 MTS/IEEE Charleston, pp. 1–7 (2018). IEEE

    2018

  45. [45]

    In: International Conference on Learning Representations (2019)

    Baker, B., Kanitscheider, I., Markov, T., Wu, Y., Powell, G., McGrew, B., Mor- datch, I.: Emergent tool use from multi-agent autocurricula. In: International Conference on Learning Representations (2019)

    2019

  46. [46]

    OpenAI, Berner, C., Brockman, G., Chan, B., Cheung, V., Dębiak, P., Dennison, C., Farhi, D., Fischer, Q., Hashme, S., Hesse, C., Józefowicz, R., Gray, S., Olsson, C., Pachocki, J., Petrov, M., Oliveira Pinto, H.P., Raiman, J., Salimans, T., Schlatter, J., Schneider, J., Sidor, S., Sutskever, I., Tang, J., Wolski, F., Zhang, S.: Dota 2 with large scale dee...

    Pith/arXiv arXiv 2019

  47. [47]

    Science Robotics7(69), 0235 (2022)

    Liu, S., Lever, G., Wang, Z., Merel, J., Eslami, S.A., Hennes, D., Czarnecki, 32 W.M., Tassa, Y., Omidshafiei, S., Abdolmaleki, A.,et al.: From motor control to team play in simulated humanoid football. Science Robotics7(69), 0235 (2022)

    2022

  48. [48]

    IEEE Robotics and Automation Letters2(2), 656–663 (2017)

    Long, P., Liu, W., Pan, J.: Deep-learned collision avoidance policy for distributed multiagent navigation. IEEE Robotics and Automation Letters2(2), 656–663 (2017)

    2017

  49. [49]

    Robot motion planning in learned latent spaces,

    Sartoretti, G., Kerr, J., Shi, Y., Wagner, G., Kumar, T.K.S., Koenig, S., Choset, H.: PRIMAL: Pathfinding via Reinforcement and Imitation Multi-Agent Learn- ing. IEEE Robotics and Automation Letters4(3), 2378–2385 (2019) https: //doi.org/10.1109/LRA.2019.2903261 . Accessed 2024-03-14

    work page doi:10.1109/lra.2019.2903261 2019

  50. [50]

    In: Proceedings of the 6th International Conference on Learning Representations

    Bansal, T., Pachocki, J., Sidor, S., Sutskever, I., Mordatch, I.: EMERGENT COMPLEXITY VIA MULTI-AGENT COMPETITION. In: Proceedings of the 6th International Conference on Learning Representations. Conference Track Proceedings, Vancouver, BC, Canada (2018)

    2018

  51. [51]

    In: Thirty- fifth Conference on Neural Information Processing Systems Datasets and Bench- marks Track (Round 1) (2021).https://openreview.net/forum?id=VdvDlnnjzIN

    Freeman, C.D., Frey, E., Raichuk, A., Girgin, S., Mordatch, I., Bachem, O.: Brax - a differentiable physics engine for large scale rigid body simulation. In: Thirty- fifth Conference on Neural Information Processing Systems Datasets and Bench- marks Track (Round 1) (2021).https://openreview.net/forum?id=VdvDlnnjzIN

    2021

  52. [52]

    arXiv preprint arXiv:2203.00806 (2022)

    Howell, T., Le Cleac’h, S., Bruedigam, J., Kolter, Z., Schwager, M., Manchester, Z.: Dojo: A differentiable simulator for robotics. arXiv preprint arXiv:2203.00806 (2022)

    arXiv 2022

  53. [53]

    ICLR (2022)

    Ren, J., Yu, C., Chen, S., Ma, X., Pan, L., Liu, Z.: Diffmimic: Efficient motion mimicking with differentiable physics. ICLR (2022)

    2022

  54. [54]

    Nature (2025) https://doi.org/10.1038/ s41586-025-08744-2

    Hafner, D., Pasukonis, J., Ba, J., Lillicrap, T.: Mastering diverse con- trol tasks through world models. Nature (2025) https://doi.org/10.1038/ s41586-025-08744-2

    2025

  55. [55]

    In: Conference on Robot Learning, pp

    Wu, P., Escontrela, A., Hafner, D., Abbeel, P., Goldberg, K.: Daydreamer: World models for physical robot learning. In: Conference on Robot Learning, pp. 2226– 2240 (2023). PMLR

    2023

  56. [56]

    In: ICML (2022)

    Hansen, N., Wang, X., Su, H.: Temporal difference learning for model predictive control. In: ICML (2022)

    2022

  57. [57]

    arXiv preprint arXiv:2501.10100 (2025)

    Li, C., Krause, A., Hutter, M.: Robotic world model: A neural network simula- tor for robust policy optimization in robotics. arXiv preprint arXiv:2501.10100 (2025)

    arXiv 2025

  58. [58]

    In: 8th Annual Conference on Robot Learning (2024)

    Song, Y., Kim, S., Scaramuzza, D.: Learning quadruped locomotion using dif- ferentiable simulation. In: 8th Annual Conference on Robot Learning (2024). https://openreview.net/forum?id=XopATjibyz 33

    2024

  59. [59]

    In: 2023 IEEE Inter- national Conference on Robotics and Automation (ICRA), pp

    Wiedemann, N., Wüest, V., Loquercio, A., Müller, M., Floreano, D., Scaramuzza, D.: Training efficient controllers via analytic policy gradient. In: 2023 IEEE Inter- national Conference on Robotics and Automation (ICRA), pp. 1349–1356 (2023). IEEE

    2023

  60. [60]

    Nature Machine Intelligence, 1–13 (2025)

    Zhang, Y., Hu, Y., Song, Y., Zou, D., Lin, W.: Learning vision-based agile flight via differentiable physics. Nature Machine Intelligence, 1–13 (2025)

    2025

  61. [61]

    arXiv preprint arXiv:1707.06347 (2017)

    Schulman, J., Wolski, F., Dhariwal, P., Radford, A., Klimov, O.: Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347 (2017)

    Pith/arXiv arXiv 2017

  62. [62]

    Makoviychuk, V., Wawrzyniak, L., Guo, Y., Lu, M., Storey, K., Macklin, M., Hoeller, D., Rudin, N., Allshire, A., Handa, A., State, G.: Isaac Gym: High Performance GPU-Based Physics Simulation For Robot Learning (2021)

    2021

  63. [63]

    In: Bengio, Y., LeCun, Y

    Kingma, D.P., Ba, J.: Adam: A method for stochastic optimization. In: Bengio, Y., LeCun, Y. (eds.) 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings (2015).http://arxiv.org/abs/1412.6980

    Pith/arXiv arXiv 2015

  64. [64]

    MIT Press, Cambridge, Mass

    Raibert, M.H.: Legged Robots that Balance. MIT Press, Cambridge, Mass. (1986) 34 Supplementary Materials Table of Contents The supplementary information in this document includes: Supplementary Sections 1-3 Supplementary Tables 1-2 Other supplementary information includes: Supplementary Video 1 Supplementary Tables Table 1: Training Hyperparameters for Lo...

    1986