Pith. sign in

REVIEW 3 major objections 5 minor 26 cited by

MuJoCo Playground

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read MuJoCo Playground is an open-source robot-learning stack that trains policies in minutes on one GPU and transfers them to real robots without fine-tuning.

desk verdict A genuinely useful open-source framework with real sim-to-real demos, where the 'zero-shot' label slightly overreaches the small trial counts. read the letter →

arxiv 2502.08844 v1 pith:F5WE4AWY submitted 2025-02-12 cs.RO

Kevin Zakka , Baruch Tabanpour , Qiayuan Liao , Mustafa Haiderbhai , Samuel Holt , Jing Yuan Luo , Arthur Allshire , Erik Frey
show 5 more authors
Koushil Sreenath Lueder A. Kahrs Carmelo Sferrazza Yuval Tassa Pieter Abbeel
This is my paper · ORCID
classification cs.RO
keywords sim-to-realtransferzero-shotpolicydeploymentGPU-acceleratedreinforcementlearningMJXbatchrenderingleggedlocomotiondexterousmanipulationvision-basedrobot
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 introduces MuJoCo Playground, a fully open-source framework for robot learning built on MJX, the JAX-based branch of the MuJoCo physics engine that runs on graphics cards. Its central claim is that pairing GPU physics with on-device batch rendering and bundled training environments lets a researcher train reinforcement-learning policies in minutes on a single GPU and deploy them directly onto real hardware, with no fine-tuning step. The authors demonstrate zero-shot sim-to-real transfer on six platforms — the Unitree Go1 quadruped, Berkeley Humanoid, Unitree G1 and Booster T1 humanoids, the LEAP hand, and the Franka arm — and report completing the whole set of deployments in under eight weeks. If the claim holds, robot learning becomes an interactive loop of train, deploy, watch, and retrain, rather than a multi-day, multi-host endeavor.

What carries the argument

The load-bearing object is the integrated stack of three components. MJX (MuJoCo XLA) is a JAX rewrite of the MuJoCo physics engine that keeps physics on the GPU, trading the dynamic memory allocation of the original engine for static shapes compiled at trace time. The Madrona batch renderer is a GPU entity-component-system renderer whose CUDA ray-tracing backend produces images on the same device as physics and learning, so vision-based policies are trained end-to-end from pixels with no teacher-student distillation. Around these, a set of environments built on MuJoCo Menagerie assets supplies the training tasks. The transfer recipe that carries the experiments is domain randomization, over sensor noise, dynamics parameters, lighting, camera pose, and object colors, combined with stochastic action and observation delays, progressive curriculum learning, and, on the arm tasks, direct high-frequency torque control at 200 Hz.

What would settle it

Retrain the pixel-based pick policy with the same ten-minute, single-GPU recipe but without the documented simplifications: allow motion in all three spatial dimensions, replace the black background and white tape with natural clutter and no positional marker, and re-enable the full collision model; then deploy the policy zero-shot and count successes over 12 trials. A collapse from the reported 12-of-12 success would show that the simplifications, not the pipeline, carried the transfer.

Watch

Extended reading notes

Core claim

The paper aims to show that GPU-based robot learning does not require closed-source simulation infrastructure. Its central claim is that MJX, a JAX implementation of MuJoCo that runs batched physics on the GPU, combined with the Madrona batch renderer, which produces pixel observations on the same device, is a sufficient stack for end-to-end sim-to-real learning. On this stack the authors train state-based policies that transfer zero-shot to four legged robots and a dexterous hand, including a LEAP-hand in-hand cube reorientation policy that trains in about 30 minutes on two RTX 4090 GPUs. They also train pixel-based policies that transfer zero-shot to a Franka arm: a non-prehensile block-reorientation policy trained with 200 Hz direct torque control reaches a median 100% and mean 85.7% success over 35 physical trials, and a 64x64 RGB pick-and-place policy trained in ten minutes on a single RTX 4090 achieves 12-of-12 real-world successes. The authors read these results as evidence that an open-source pipeline can deliver the kinds of sim-to-real results previously associated with closed-source GPU simulators.

Load-bearing premise

The whole framework's zero-shot sim-to-real claim rests on the assumption that the deliberate task simplifications — the fixed Y-Z plane, black background, white tape over the cube's range, and disabled collisions in the pixel-based pick task — are not what enables the real-world success, so that the transfer would survive without them.

Editorial extensions

If this is right

  • Contact-rich tasks such as in-hand cube reorientation train in roughly 30 minutes on two consumer GPUs, so reward prototyping becomes an interactive process rather than an overnight batch job.
  • Vision-based policies can be trained directly from pixels with physics, rendering, and learning all on-device, removing the distillation step that earlier pixel-based sim-to-real pipelines required.
  • Because the physics engine, renderer, and environments are all open source, researchers can modify the simulation internals for their own tasks, which is not possible with the closed-source GPU physics pipelines the paper contrasts against.
  • The training bottleneck shifts from data collection to policy-network updates: physics, rendering, and inference together amount to only 9% of total training time for the Cartpole pixel task and 43% for the Franka pick task.

Reading between the lines

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

  • The paper's own timing breakdown implies that future speedups for pixel-based training will come from cheaper vision architectures or more sample-efficient algorithms, since policy updates now dominate wall-clock time.
  • The pixel-pick result is demonstrated under conditions the paper discloses openly — a fixed Y-Z plane, a black background, white tape marking the cube's possible range, and collisions disabled except between gripper fingers and cube — so treating zero-shot transfer from pixels as established for full 3D, cluttered manipulation is an extrapolation the paper does not test.
  • The same training recipe transferring across four legged morphologies and three manipulation setups in under eight weeks suggests the framework's main contribution is reproducibility and iteration speed; the paper itself does not claim algorithmic novelty.
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 / 5 minor

Summary. The paper introduces MuJoCo Playground, an open-source robot-learning framework built on MJX and the Madrona batch renderer, providing GPU-accelerated physics, on-device rendering, and a suite of training environments. It reports multi-seed training curves and throughput measurements for DM Control Suite, locomotion, and manipulation tasks, and claims zero-shot sim-to-real transfer on six platforms from both state and pixel observations: LEAP hand cube reorientation (10 trials), Franka non-prehensile block reorientation (35 trials), Franka pixel pick-cube (12 trials), and qualitative deployments on Unitree Go1, Berkeley Humanoid, Unitree G1, and Booster T1. The paper also provides hyperparameters, notebooks, and code as part of the open-source release.

Significance. If the reported results hold, this is a valuable open-source infrastructure contribution: it demonstrates a pip-installable, Colab-compatible training stack with credible training curves, measured throughput across consumer and datacenter GPUs, and an integrated batch renderer that supports end-to-end pixel-based RL without teacher-student distillation. The throughput comparisons against IsaacLab and ManiSkill3 are useful even though the authors appropriately describe them as rough. The strongest assets are the reproducibility-oriented release: environment code, training curves, hyperparameters, and notebooks accompany the paper. My assessment is that the central engineering claims are sound, but the breadth of the zero-shot sim-to-real claim is not fully supported by the reported evidence, particularly for pixel-based policies; this is the load-bearing issue that drives my recommendation.

major comments (3)
  1. [Abstract; Section V.b; Tables I-II; Section IV.B] The abstract and related-work section claim zero-shot sim-to-real transfer across six platforms without caveat, but the quantitative evidence per platform is thin: 10 trials for the LEAP hand with a median of only 3.5 consecutive rotations before failure (Table I), 12 trials for the pixel pick-cube (Section IV.C.3.d), 35 trials for the non-prehensile task (Table II), and no trial counts or quantitative metrics for the four locomotion deployments (Sections IV.B.1.d and IV.B.2.d, which refer only to videos and qualitative robustness). Please either provide per-platform metrics with trial counts and confidence intervals for all six platforms, or restrict the zero-shot claim to the demonstrated tasks and state explicitly that the locomotion evidence is qualitative.
  2. [Section C.6; Section IV.C.3.d] The claim that the pick-cube result demonstrates 'capacity for training pixel-based policies that transfer reliably' is broader than what the evidence supports. The real task is heavily constrained: the end-effector is restricted to a fixed Y-Z plane, the block range is only 20 cm, the background is black, the policy receives a white-tape cue over the possible starting positions, and collisions are disabled except between gripper fingers and cube. With 12 trials and a 100% success rate, the experiment cannot rule out that the white-tape cue and the collision simplifications are the main enablers of transfer. Please report ablations, or at minimum the policy's success without the tape cue and with fuller collision geometry, and state these simplifications prominently wherever the pixel-transfer capability claim is made.
  3. [Section IV.B.1.d; Section IV.B.2.d] The locomotion results make up half of the six-platform breadth claim, yet they are reported only as qualitative video demonstrations. There are no success criteria, trial counts, commanded-versus-achieved velocity tracking errors, or perturbation protocols for the Go1, Berkeley Humanoid, G1, or Booster T1 deployments. Since the paper's central claim is zero-shot transfer across platforms, this is a substantive evidence gap. Please add quantitative locomotion metrics with trial counts, or explicitly mark the locomotion results as preliminary demonstrations rather than validated zero-shot results.
minor comments (5)
  1. [Section IV.B.1.c] The sentence 'We firstly train the policy...' should be 'We first train the policy...'.
  2. [Section V.c] The text contains a typo: 'As as result' should be 'As a result'.
  3. [Figure 8 caption] The caption uses lowercase 'brax' for the library name; it should be capitalized as 'Brax' for consistency with the text.
  4. [Section C.4.41a] The phrase 'as opposed to 0.4 rad in the real-world setup' is confusing because the preceding number is the simulation tolerance; please rephrase to make clear that the simulation uses 0.1 rad and the real-world evaluation uses 0.4 rad.
  5. [Section VI; Section IV.C.3.d] The Limitations section appropriately states that 'vision-based training using Madrona is still at an early stage'; this caveat should be echoed in the main-text sentence that highlights the pick-cube result, so readers do not overgeneralize the pixel-transfer claim.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the paper's claims are direct measurements of its own open-source framework and real-hardware deployments; self-citations are infrastructural, code-reproduced tools and are not used to justify the zero-shot transfer results.

full rationale

This is an empirical systems paper with no derivation chain to audit: there is no equation from which a result is derived, no fitted parameter renamed as a prediction, and no uniqueness theorem invoked. The central claims (rapid training on a single GPU; zero-shot sim-to-real on six platforms) are supported by measurements the paper itself reports: training wallclock times, throughput tables, and real-hardware trial counts (Table I and Table II in Section IV-C, and the 12-trial pick-cube run in Section C.6.d). Those real-world deployments are externally falsifiable evidence and are not equivalent to the paper's inputs by construction: the simulator, rewards, and domain randomization do not guarantee the reported hardware successes, which the paper presents as observed outcomes with honest caveats. The self-citations to MJX [43], MuJoCo Menagerie [68], Brax [13], MuJoCo [63], and the DM Control Suite [61] are infrastructural -- the framework is literally built on these open-source, code-reproduced tools -- and they are not used to justify the zero-shot transfer results, which rest on the paper's own experiments. Per the review rubric, code-reproduced, externally verifiable artifacts do not raise the circularity score. The system-identification constants in Section C.42 (armature inertia computed from the supplier data sheet; friction set to 10% of maximum torque) are stated assumptions hedged by heavy domain randomization, not predictions fitted to the deployment data. Disclosed task simplifications (fixed Y-Z plane and white tape in Section C.6; disabling all but gripper-cube collisions) qualify the breadth of the pixel-based transfer claim but do not make the claim a restatement of its own inputs, and the paper explicitly flags that vision-based training via Madrona is 'still at an early stage' (Section VI). Concerns that trial counts are small or that locomotion results are qualitative are evidence-adequacy issues, not circularity, and are out of scope for this pass. The honest non-finding is therefore a score of 0.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The central claims of the paper rest on a set of hand-tuned parameters (reward weights, delays, noise ranges, torque limits) and on the domain assumptions that MJX physics, Madrona rendering, and domain randomization together approximate the real world closely enough for zero-shot transfer. No new physical entities are introduced.

free parameters (6)
  • Leap hand armature inertia (Ia) = 1.41e-3 kg m^2
    Computed from kg=288.35, mr=2.0e-3 kg, rr=4.12e-3 m in Section C.42; used in the simulated motor model and load-bearing for the dexterous hand transfer.
  • Nominal actuator friction = 10% of maximum torque
    Ad hoc value chosen in Section C.42 because friction is hard to measure; the policy is trained with heavy randomization around it.
  • Cube pose observation noise = 2 cm position, 0.1 rad rotation
    Added in Section C.41b to mimic pose estimator noise; hand-selected.
  • Action/observation delay ranges = actions 1-3 steps, observations 6-12 steps
    Used in Section C.51c to mimic hardware latency; chosen by engineering judgment, not measured.
  • Torque limit for block reorientation = 8 Nm per joint
    Set in Section C.51c as approximate safe limit for the Franka arm; affects policy behavior.
  • Joystick command ranges = train {1.5, 0.8, 1.2}, finetune {1.5, 0.8, 2π}
    Chosen in Section B.25a to balance exploration; affects gait quality.
assumptions (5)
  • domain assumption MJX simulation is a sufficiently faithful model of real-world rigid-body dynamics for the tasks used
    The entire sim-to-real pipeline rests on this; introduced in Sections II and IV.
  • domain assumption Domain randomization adequately covers the sim-to-real gap for each hardware platform
    Stated as the method for transfer in Sections IV.B and IV.C.
  • domain assumption The simplified vision task (fixed Y-Z plane, black background, tape marker) is a valid proxy for vision-based robot learning
    Section C.6 describes these simplifications as enabling robust deployment, but the paper generalizes to zero-shot sim-to-real from pixel inputs in the abstract.
  • domain assumption The LEAP hand actuator model with assumed rotor inertia and 10% friction is accurate enough for policy transfer
    Section C.42 performs system identification using supplier data and an assumed rotor mass distribution; this is necessary for the in-hand results.
  • domain assumption The small trial counts (n=10 to 35) are representative of true real-world success rates
    The paper reports success rates from a handful of trials without statistical power analysis; e.g., Table I and Section C.6d.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MuJoCo Playground." pith.science (2026). https://pith.science/paper/F5WE4AWY

@misc{pith2026250208844,
  author       = {Pith},
  title        = {Pith review of: MuJoCo Playground},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/F5WE4AWY}},
  note         = {Machine review of arXiv:2502.08844}
}
read the original abstract

We introduce MuJoCo Playground, a fully open-source framework for robot learning built with MJX, with the express goal of streamlining simulation, training, and sim-to-real transfer onto robots. With a simple "pip install playground", researchers can train policies in minutes on a single GPU. Playground supports diverse robotic platforms, including quadrupeds, humanoids, dexterous hands, and robotic arms, enabling zero-shot sim-to-real transfer from both state and pixel inputs. This is achieved through an integrated stack comprising a physics engine, batch renderer, and training environments. Along with video results, the entire framework is freely available at playground.mujoco.org

Figures

Figures reproduced from arXiv: 2502.08844 by the authors.

Figure 1
Figure 1. A cartoon of MuJoCo Playground’s diverse environments that were successfully transferred to real hardware, including Berkeley Humanoid, Unitree Go1 and G1, LEAP hand and Franka Arm. Abstract—We introduce MuJoCo Playground, a fully open￾source framework for robot learning built with MJX, with the express goal of streamlining simulation, training, and sim￾to-real transfer onto robots. With a simple pip install playgro… view at source ↗
Figure 2
Figure 2. A preview of locomotion and manipulation environments available in MuJoCo Playground. III. BATCH RENDERING WITH MADRONA MuJoCo Playground enables vision-based environments through an integration of MJX with Madrona [54]. Madrona is a GPU-based entity-component-system (ECS), which contains GPU implementations of high throughput rendering [49]. Madrona provides two rendering backends: a software-based batch ray tracer… view at source ↗
Figure 3
Figure 3. Several DM Control Suite environments [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (21 more)
Figure 4
Figure 4. Figure 4: Sample renders from the Madrona batch renderer for the Panda and Aloha environments. Left-most images are the original environments. The remaining images highlight the the support for lighting, shadows, textures, and colors, including the ability to domain randomize th…
Figure 5
Figure 5. Figure 5: Footage from four of our deployed policies. a) Go1 joystick policy recovering from a kick while travelling at ∼ 2m/s, b) Berkeley humanoid joystick policy tracking an angular velocity command on a slippery surface. c) In-Hand Cube Reorientation transitioning between tw…
Figure 6
Figure 6. Figure 6: Training wallclock time for LeapCubeReorient on different GPU device topologies. 1x 4090 takes ∼ 2080 (s) to train and 8x H100 takes ∼ 670 (s) to train. All runs use the same hyperparams (e.g. 8192 num envs); we leave tuning hyperparams per topology as a future exercis…
Figure 7
Figure 7. Figure 7: Environment steps per second on the single-camera Cartpole￾Balance and PandaPickCubeCartesian environments with pixel-based observations from our on-device renderer. in the appendix, we report RL training throughput for all environments in MuJoCo Playground on a single…
Figure 8
Figure 8. Figure 8: Reward curves for PPO trained with RSL-RL and brax on an RTX-4090 GPU for 3 seeds each on the Unitree Go1. networks. Hence, compared to traditional on-policy training pipelines, we have shifted our bottleneck from collecting data to processing it. Training bottlenecks …
Figure 9
Figure 9. Figure 9: Reward vs environment steps for PPO and SAC on the full DM Control Suite environments in MuJoCo Playground. We run PPO for 60M steps, with a few selected environments running on 100M steps. SAC runs for 5M steps. All settings are run with 5 seeds on a single A100 GPU d…
Figure 10
Figure 10. Figure 10: Reward vs wallclock time for PPO and SAC on the full DM Control Suite environments in MuJoCo Playground. All settings are run with 5 seeds on a single A100 GPU device [PITH_FULL_IMAGE:figures/full_fig_p015_10.png]
Figure 11
Figure 11. Figure 11: Reward vs environment steps for Brax PPO. All settings are run with 5 seeds on a single A100 GPU device [PITH_FULL_IMAGE:figures/full_fig_p019_11.png]
Figure 12
Figure 12. Figure 12: Reward vs wallclock time for Brax PPO. All settings are run with 5 seeds on a single A100 GPU device. Notice that the initial flat region measures the compilation time for the training + environment code [PITH_FULL_IMAGE:figures/full_fig_p020_12.png]
Figure 13
Figure 13. Figure 13: Training wallclock time for Go1JoystickFlatTerrain on different GPU devices and topologies [PITH_FULL_IMAGE:figures/full_fig_p021_13.png]
Figure 14
Figure 14. Figure 14 [PITH_FULL_IMAGE:figures/full_fig_p023_14.png]
Figure 15
Figure 15. Figure 15 [PITH_FULL_IMAGE:figures/full_fig_p024_15.png]
Figure 16
Figure 16. Figure 16: Training wallclock time for LeapHandReorient on different GPU devices and topologies [PITH_FULL_IMAGE:figures/full_fig_p025_16.png]
Figure 17
Figure 17. Figure 17: Example MuJoCo scene of our block reorientation environment. The block is pushed toward the center. C.5. Real-world Non-prehensile Block Reorientation with a Franka-Robotiq Arm In this section, we provide technical details for our block reorientation task on a real Fr…
Figure 18
Figure 18. Figure 18: Real Franka Emika Panda robot with a Robotiq gripper, pushing the yoga block to the goal region [PITH_FULL_IMAGE:figures/full_fig_p029_18.png]
Figure 19
Figure 19. Figure 19: Policy inputs across domain randomized environments (64x64 pixels each) used while training the deployed PandaPickCubeCartesian agent. Lighting conditions, colors, brightness and camera pose are all randomized. C.6. Real-world Franka PickCube from Pixels To highlight …
Figure 20
Figure 20. Figure 20: Left. Franka Research robot with a Realsense camera capturing input images. Right. Policy inputs from one embodied rollout [PITH_FULL_IMAGE:figures/full_fig_p031_20.png]
Figure 21
Figure 21. Figure 21: Reward vs environment steps for brax PPO. All settings are run with 5 seeds on a single RTX 4090 GPU. D.2. Performance Benchmarking In this section we benchmark the througput of Madrona MJX GPU batch rendering. For reference, we plot our results alongside those from I…
Figure 22
Figure 22. Figure 22: Reward vs wallclock time for brax PPO. All settings are run with 5 seeds on a single RTX 4090 GPU [PITH_FULL_IMAGE:figures/full_fig_p033_22.png]
Figure 23
Figure 23. Figure 23: Comparison of raw environment-stepping throughput with prior simulators for CartpoleBalance with state-based and pixel observations of varying sizes [PITH_FULL_IMAGE:figures/full_fig_p034_23.png]
Figure 24
Figure 24. Figure 24: Time-cost breakdown of unrolling physics simulation and rendering for CartpoleBalance with pixel observations. Lower is better. Per-step rendering time is stacked without overlap over physics simulation time [PITH_FULL_IMAGE:figures/full_fig_p035_24.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 26 Pith papers

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

  1. On Data Thinning for Model Validation in Small Area Estimation

    stat.ME 2026-04 unverdicted novelty 7.0 of 10

    Thinned-data MSE for small-area models is unbiased for a risk that systematically differs from full-data risk; under Fay-Herriot the gap is closed-form in the model's shrinkage, and the thinning fraction faces a sharp...

  2. Flow Matching Policy Gradients

    cs.LG 2025-07 conditional novelty 7.0 of 10

    FPO trains flow-based policies with PPO by replacing the likelihood ratio with an exponentiated flow matching loss difference.

  3. Visual Imitation Enables Contextual Humanoid Control

    cs.RO 2025-05 conditional novelty 7.0 of 10

    A single policy trained from 123 monocular videos, fine-tuned in simulation, and distilled to heightmap plus root-direction inputs lets a Unitree G1 climb stairs and sit and stand on real furniture.

  4. V-Simba: Unleashing the Architectural Potential of RL in Visual Continuous Control

    cs.LG 2026-08 conditional novelty 6.0 of 10

    V-Simba, a visual RL architecture combining layer normalization, weight decay, and a distributional critic, matches or outperforms complex baselines on 29 continuous control tasks while using less compute.

  5. ATP: Anatomical Torque with Passivity-based Control Framework for Safe Upper-Limb Exoskeleton Assistance

    cs.RO 2026-08 conditional novelty 6.0 of 10

    An upper-limb exoskeleton framework generates muscle-based torque references with reinforcement learning, refines them online, and delivers them via a passivity-preserving torque controller, with a pilot EMG study sho...

  6. LooperMuscle: Fast and Stable Learning of Humanoid Whole-Body Tracking via Structured Mixture-of-Experts

    cs.RO 2026-08 conditional novelty 6.0 of 10

    LooperMuscle trains a humanoid whole-body tracking policy in about 45 minutes, cutting body error by 34% versus FastSAC and reaching 72% of PPO's reward, versus PPO's 6 hours.

  7. $\pi\mathbf{R}^2$: Reactive Real-time Flow Policies

    cs.RO 2026-07 conditional novelty 6.0 of 10

    πR² makes flow-matching VLA policies reactive by splitting conditioning into fresh proprioception and stale vision-language features and using a one-step-per-call staircase noise schedule, reaching ~25 Hz closed-loop ...

  8. OrchardBench: A Physically-Grounded, GPU-Parallel Apple-Orchard Simulation Benchmark for Agricultural Robotics

    cs.RO 2026-07 conditional novelty 6.0 of 10

    OrchardBench simulates physically-grounded, breakable, fruit-bearing apple trees on a GPU-parallel engine to benchmark autonomous harvesting robots.

  9. Cross-Embodiment Robot Manipulation via a Unified Hand Action Space

    cs.RO 2026-07 conditional novelty 6.0 of 10

    UHAS maps hand actions to deformations of a shared unit sphere and recovers joint commands via cascade IK, enabling multi-hand RL, zero-shot transfer, and modest real-world cube reorientation on LEAP and Allegro.

  10. Research Novelty in Information Systems Journals After ChatGPT: Differences Across Institutional Language Contexts

    cs.DL 2026-03 unverdicted novelty 6.0 of 10

    Post-2022, IS articles from non-English-dominant first-author institutions show a 0.176 SD larger decline in relative semantic novelty (~7 percentile points) than English-dominant ones.

  11. PTLD: Sim-to-real Privileged Tactile Latent Distillation for Dexterous Manipulation

    cs.RO 2026-03 unverdicted novelty 6.0 of 10

    PTLD distills real privileged tactile data into a state estimator to boost sim-to-real performance of proprioceptive dexterous manipulation policies, yielding 182% improvement on in-hand rotation and 57% on reorientat...

  12. MyoInteract: A Framework for Fast Prototyping of Biomechanical HCI Tasks using Reinforcement Learning

    cs.HC 2026-02 conditional novelty 6.0 of 10

    MyoInteract makes biomechanical reinforcement-learning user simulation fast and GUI-accessible, reducing pointing-task training from ~7 hours to ~10–36 minutes while preserving Fitts'-law-compatible movements.

  13. Simultaneous Contact Sequence and Patch Planning for Dynamic Locomotion

    cs.RO 2025-08 conditional novelty 6.0 of 10

    A Monte Carlo tree search and trajectory optimization pipeline that simultaneously chooses contact sequences and contact patches for legged robots, demonstrated on a quadruped and a humanoid.

  14. SHIELD: Safety on Humanoids via CBFs In Expectation on Learned Dynamics

    cs.RO 2025-05 conditional novelty 6.0 of 10

    SHIELD learns a generative model of a humanoid's tracking error and uses a stochastic control barrier function to filter reference commands, giving runtime obstacle avoidance with probabilistic safety bounds in hardware.

  15. MOSAIC: Skill-Centric Manipulation Planning with Physics Simulation

    cs.RO 2025-04 conditional novelty 6.0 of 10

    MOSAIC is a multi-directional skill-centric planner that seeds feasible local trajectories with generator skills, links them with connector skills, and uses a statistical oracle and physics simulation to guide the search.

  16. AXIS: A Growable Community-Driven Data Engine for Scalable Robot Manipulation

    cs.RO 2026-07 conditional novelty 5.0 of 10

    Pretraining π0.5 on the crowdsourced AXIS simulation dataset (207 tasks, 50K+ trajectories) raises downstream LIBERO-Plus success from 83.9% to 88.8% as the pretraining corpus grows from none to the full dataset.

  17. What Matters for Simulation to Online Reinforcement Learning on Real Robots

    cs.RO 2026-02 conditional novelty 5.0 of 10

    Sim-to-online RL on three real robots is stabilized by retaining data, warm-starting the replay buffer, and using asymmetric actor-critic updates with a low actor learning rate.

  18. RecoWorld: Building Simulated Environments for Agentic Recommender Systems

    cs.IR 2025-09 conditional novelty 5.0 of 10

    A design proposal, not a tested system: a dual-view simulation loop in which an LLM-simulated user issues reflective instructions when about to disengage, and an instruction-following recommender adapts to maximize si...

  19. Viser: Imperative, Web-based 3D Visualization in Python

    cs.CV 2025-07 accept novelty 5.0 of 10

    The paper describes Viser, an open-source imperative, web-based 3D visualization library for Python with scene and GUI primitives.

  20. SimLauncher: Launching Sample-Efficient Real-world Robotic Reinforcement Learning via Simulation Pre-training

    cs.RO 2025-07 conditional novelty 5.0 of 10

    Simulation-pretrained policies, with digital-twin demos for critic bootstrapping and action proposals, cut real-world RL training time while reaching near-perfect success on three manipulation tasks.

  21. Foundations of Reinforcement Learning and Control:Connections and New Perspectives

    cs.LG 2026-08 conditional novelty 4.0 of 10

    A SAC-trained Half-Cheetah policy paired with a low-level model-reference adaptive controller recovers running performance after a change in joint damping, where the fixed learned policy alone fails.

  22. Rethinking the Suitability of Reinforcement Learning Algorithms Under Practical Transfer Constraints

    cs.LG 2026-07 conditional novelty 4.0 of 10

    Under wall-clock budgets, sample-inefficient PPO on 2,048 parallel environments often reached a usable policy faster than sample-efficient SAC and TD-MPC2, while domain randomization offered no paradigm-specific robus...

  23. Robotic Manipulation via Imitation Learning: Taxonomy, Evolution, Benchmark, and Challenges

    cs.RO 2025-08 conditional novelty 4.0 of 10

    A survey that taxonomizes robotic manipulation policies trained by imitation learning, traces their evolution, and compiles benchmark comparisons.

  24. Booster Gym: An End-to-End Reinforcement Learning Framework for Humanoid Robot Locomotion

    cs.RO 2025-06 conditional novelty 4.0 of 10

    Booster Gym provides an end-to-end, open-source RL pipeline with zero-shot sim-to-real transfer, validated on the Booster T1 humanoid for omnidirectional walking, terrain adaptation, and push recovery.

  25. FastTD3: Simple, Fast, and Capable Reinforcement Learning for Humanoid Control

    cs.RO 2025-05 conditional novelty 4.0 of 10

    FastTD3, a tuned TD3 with parallel environments and large-batch distributional critic updates, solves humanoid benchmark tasks in under 3 hours on one GPU.

  26. Monte Carlo Beam Search for Actor-Critic Reinforcement Learning in Continuous Control

    cs.AI 2025-05 reject novelty 4.0 of 10

    A TD3 variant that evaluates multiple perturbed actions via short Monte Carlo rollouts reports faster learning and higher returns on HalfCheetah, Walker2d, and Swimmer.

Reference graph

Works this paper leans on

78 extracted references · 40 canonical work pages · cited by 26 Pith papers

  1. [1]

    Legged locomotion in challenging ter- rains using egocentric vision

    Ananye Agarwal, Ashish Kumar, Jitendra Malik, and Deepak Pathak. Legged locomotion in challenging ter- rains using egocentric vision. In Conference on robot learning, pages 403–415. PMLR, 2023

    work page 2023

  2. [2]

    Aloha 2: An enhanced low- cost hardware for bimanual teleoperation

    Jorge ALOHA 2 Team, Aldaco, Travis Armstrong, Robert Baruch, Jeff Bingham, Sanky Chan, Kenneth Draper, Debidatta Dwibedi, Chelsea Finn, Pete Florence, Spencer Goodrich, et al. Aloha 2: An enhanced low- cost hardware for bimanual teleoperation. arXiv preprint arXiv:2405.02292, 2024

    arXiv 2024

  3. [3]

    What matters in on-policy reinforcement learning? a large-scale empirical study

    Marcin Andrychowicz, Anton Raichuk, Piotr Sta ´nczyk, Manu Orsini, Sertan Girgin, Raphael Marinier, L ´eonard Hussenot, Matthieu Geist, Olivier Pietquin, Marcin Michalski, et al. What matters in on-policy reinforcement learning? a large-scale empirical study. arXiv preprint arXiv:2006.05990, 2020

    arXiv 2006

  4. [4]

    Learning dexterous in-hand manipula- tion

    OpenAI: Marcin Andrychowicz, Bowen Baker, Maciek Chociej, Rafal Jozefowicz, Bob McGrew, Jakub Pa- chocki, Arthur Petron, Matthias Plappert, Glenn Powell, Alex Ray, et al. Learning dexterous in-hand manipula- tion. The International Journal of Robotics Research , 39 (1):3–20, 2020

    work page 2020

  5. [5]

    JAX: composable transformations of Python+NumPy programs, 2018

    James Bradbury, Roy Frostig, Peter Hawkins, Matthew James Johnson, Chris Leary, Dougal Maclaurin, George Necula, Adam Paszke, Jake VanderPlas, Skye Wanderman-Milne, and Qiao Zhang. JAX: composable transformations of Python+NumPy programs, 2018. URL http://github.com/jax-ml/jax

    2018

  6. [6]

    Bark- our: Benchmarking animal-level agility with quadruped robots

    Ken Caluwaerts, Atil Iscen, J Chase Kew, Wenhao Yu, Tingnan Zhang, Daniel Freeman, Kuang-Huei Lee, Lisa Lee, Stefano Saliceti, Vincent Zhuang, et al. Bark- our: Benchmarking animal-level agility with quadruped robots. arXiv preprint arXiv:2305.14654 , 2023

    arXiv 2023

  7. [7]

    Closing the sim-to-real loop: Adapting simulation randomization with real world experience

    Yevgen Chebotar, Ankur Handa, Viktor Makoviychuk, Miles Macklin, Jan Issac, Nathan Ratliff, and Dieter Fox. Closing the sim-to-real loop: Adapting simulation randomization with real world experience. In 2019 International Conference on Robotics and Automation (ICRA), pages 8973–8979. IEEE, 2019

    2019

  8. [8]

    A system for general in-hand object re-orientation

    Tao Chen, Jie Xu, and Pulkit Agrawal. A system for general in-hand object re-orientation. In Conference on Robot Learning, pages 297–307. PMLR, 2022

    2022

Show all 78 references
  1. [9]

    Extreme parkour with legged robots

    Xuxin Cheng, Kexin Shi, Ananye Agarwal, and Deepak Pathak. Extreme parkour with legged robots. In 2024 IEEE International Conference on Robotics and Automa- tion (ICRA), pages 11443–11450. IEEE, 2024

    2024

  2. [10]

    Corn: Contact-based object represen- tation for nonprehensile manipulation of general unseen objects

    Yoonyoung Cho, Junhyek Han, Yoontae Cho, and Beomjoon Kim. Corn: Contact-based object represen- tation for nonprehensile manipulation of general unseen objects. arXiv preprint arXiv:2403.10760 , 2024

    2024 arXiv

  3. [11]

    Onnx runtime

    ONNX Runtime developers. Onnx runtime. https: //onnxruntime.ai/, 2021. Version: x.y.z

    2021

  4. [12]

    Flayols, A

    T. Flayols, A. Del Prete, P. Wensing, A. Mifsud, M. Be- nallegue, and O. Stasse. Experimental evaluation of sim- ple estimators for humanoid robots. In 2017 IEEE-RAS 17th International Conference on Humanoid Robotics (Humanoids), pages 889–895, 2017. doi: 10.1109/ HUMANOIDS.20...

    2017

  5. [13]

    Brax-a differentiable physics engine for large scale rigid body simulation, 2021

    C Daniel Freeman, Erik Frey, Anton Raichuk, Sertan Girgin, Igor Mordatch, and Olivier Bachem. Brax-a differentiable physics engine for large scale rigid body simulation, 2021. URL http://github. com/google/brax, 6, 2021

    2021

  6. [14]

    Genesis: A universal and generative physics engine for robotics and beyond, December 2024

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

    2024

  7. [15]

    Soft actor-critic: Off-policy maximum entropy deep reinforcement learning with a stochastic actor

    Tuomas Haarnoja, Aurick Zhou, Pieter Abbeel, and Sergey Levine. Soft actor-critic: Off-policy maximum entropy deep reinforcement learning with a stochastic actor. In International conference on machine learning , pages 1861–1870. PMLR, 2018

    2018

  8. [16]

    Learning agile soccer skills for a bipedal robot with deep reinforcement learning

    Tuomas Haarnoja, Ben Moran, Guy Lever, Sandy H Huang, Dhruva Tirumala, Jan Humplik, Markus Wulfmeier, Saran Tunyasuvunakool, Noah Y Siegel, Roland Hafner, et al. Learning agile soccer skills for a bipedal robot with deep reinforcement learning. Science Robotics, 9(89):eadi8022, 2024

    2024

  9. [17]

    Mastering diverse domains through world models

    Danijar Hafner, Jurgis Pasukonis, Jimmy Ba, and Timo- thy Lillicrap. Mastering diverse domains through world models. arXiv preprint arXiv:2301.04104 , 2023

    2023 arXiv

  10. [18]

    Mustafa Haiderbhai, Radian Gondokaryono, Andrew Wu, and Lueder A. Kahrs. Sim2real rope cutting with a surgical robot using vision-based reinforcement learning. Transactions on Automation Science and Engineering ,

  11. [19]

    Dextreme: Transfer of agile in-hand manipulation from simulation to reality

    Ankur Handa, Arthur Allshire, Viktor Makoviychuk, Aleksei Petrenko, Ritvik Singh, Jingzhou Liu, Denys Makoviichuk, Karl Van Wyk, Alexander Zhurkevich, Balakumar Sundaralingam, et al. Dextreme: Transfer of agile in-hand manipulation from simulation to reality. In 2023 IEEE Inte...

    2023

  12. [20]

    Td-mpc2: Scalable, robust world models for continuous control, 2024

    Nicklas Hansen, Hao Su, and Xiaolong Wang. Td-mpc2: Scalable, robust world models for continuous control, 2024

    2024

  13. [21]

    Analytical inverse kinematics for franka emika panda – a geometrical solver for 7- dof manipulators with unconventional design

    Yanhao He and Steven Liu. Analytical inverse kinematics for franka emika panda – a geometrical solver for 7- dof manipulators with unconventional design. In 2021 9th International Conference on Control, Mechatronics and Automation (ICCMA) , pages 194–199, 2021. doi: 10.1109/IC...

    2021

  14. [22]

    Evolving control: Evolved high frequency control for continuous control tasks

    Samuel Holt, Todor Davchev, Dhruva Tirumala, Ben Moran, Yixin Lin, Antoine Laurens, Atil Iscen, Erik Frey, Markus Wulfmeier, Francesco Romano, and Nico- las Heess. Evolving control: Evolved high frequency control for continuous control tasks. In CoRL Workshop on Safe and Robus...

    2024

  15. [23]

    Difftaichi: Differentiable programming for physical sim- ulation

    Yuanming Hu, Luke Anderson, Tzu-Mao Li, Qi Sun, Nathan Carr, Jonathan Ragan-Kelley, and Fr ´edo Durand. Difftaichi: Differentiable programming for physical sim- ulation. arXiv preprint arXiv:1910.00935 , 2019

    1910 arXiv

  16. [24]

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

    Julian Ibarz, Jie Tan, Chelsea Finn, Mrinal Kalakrishnan, Peter Pastor, and Sergey Levine. How to train your robot with deep reinforcement learning: lessons we have learned. The International Journal of Robotics Research, 40(4-5):698–721, 2021

    2021

  17. [25]

    Concurrent training of a control policy and a state estimator for dynamic and robust legged locomotion

    Gwanghyeon Ji, Juhyeok Mun, Hyeongjun Kim, and Jemin Hwangbo. Concurrent training of a control policy and a state estimator for dynamic and robust legged locomotion. IEEE Robotics and Automation Letters , 7 (2):4630–4637, April 2022. ISSN 2377-3774. doi: 10. 1109/lra.2022.3151...

    2022

  18. [26]

    Champion-level drone racing using deep rein- forcement learning

    Elia Kaufmann, Leonard Bauersfeld, Antonio Loquercio, Matthias M ¨uller, Vladlen Koltun, and Davide Scara- muzza. Champion-level drone racing using deep rein- forcement learning. Nature, 620(7976):982–987, 2023

    2023

  19. [27]

    Reinforce- ment learning in robotics: A survey

    Jens Kober, J Andrew Bagnell, and Jan Peters. Reinforce- ment learning in robotics: A survey. The International Journal of Robotics Research , 32(11):1238–1274, 2013

    2013

  20. [28]

    Design and use paradigms for gazebo, an open-source multi-robot sim- ulator

    Nathan Koenig and Andrew Howard. Design and use paradigms for gazebo, an open-source multi-robot sim- ulator. In 2004 IEEE/RSJ international conference on intelligent robots and systems (IROS)(IEEE Cat. No. 04CH37566), volume 3, pages 2149–2154. Ieee, 2004

    2004

  21. [29]

    Reinforcement learning with augmented data

    Michael Laskin, Kimin Lee, Adam Stooke, Lerrel Pinto, Pieter Abbeel, and Aravind Srinivas. Reinforcement learning with augmented data. arXiv:2004.14990

    2004 arXiv

  22. [30]

    Robust recovery controller for a quadrupedal robot using deep re- inforcement learning

    Joonho Lee, Jemin Hwangbo, and Marco Hutter. Robust recovery controller for a quadrupedal robot using deep re- inforcement learning. arXiv preprint arXiv:1901.07517 , 2019

    1901 arXiv

  23. [31]

    rsl rl: Fast and simple implementation of rl algorithms, designed to run fully on gpu

    leggedrobotics. rsl rl: Fast and simple implementation of rl algorithms, designed to run fully on gpu. https://github. com/leggedrobotics/rsl rl, 2023. Accessed: January 10, 2025

    2023

  24. [32]

    Albert Hao Li, Preston Culbertson, Vince Kurtz, and Aaron D. Ames. Drop: Dexterous reorientation via online planning. arXiv preprint arXiv:2409.14562, 2024. Available at: https://arxiv.org/abs/2409.14562

    2024 arXiv

  25. [33]

    Rein- forcement learning for versatile, dynamic, and robust bipedal locomotion control

    Zhongyu Li, Xue Bin Peng, Pieter Abbeel, Sergey Levine, Glen Berseth, and Koushil Sreenath. Rein- forcement learning for versatile, dynamic, and robust bipedal locomotion control. The International Journal of Robotics Research , page 02783649241285161, 2024

    2024

  26. [34]

    Gpu- accelerated robotic simulation for distributed reinforce- ment learning

    Jacky Liang, Viktor Makoviychuk, Ankur Handa, Nut- tapong Chentanez, Miles Macklin, and Dieter Fox. Gpu- accelerated robotic simulation for distributed reinforce- ment learning. In Conference on Robot Learning , pages 270–282. PMLR, 2018

    2018

  27. [35]

    Berkeley hu- manoid: A research platform for learning-based control

    Qiayuan Liao, Bike Zhang, Xuanyu Huang, Xiaoyu Huang, Zhongyu Li, and Koushil Sreenath. Berkeley hu- manoid: A research platform for learning-based control. arXiv preprint arXiv:2407.21781 , 2024

    2024 arXiv

  28. [36]

    Learning hu- manoid locomotion with perceptive internal model

    Junfeng Long, Junli Ren, Moji Shi, Zirui Wang, Tao Huang, Ping Luo, and Jiangmiao Pang. Learning hu- manoid locomotion with perceptive internal model. arXiv preprint arXiv:2411.14386, 2024

    2024 arXiv

  29. [37]

    Eureka: Human- level reward design via coding large language models

    Yecheng Jason Ma, William Liang, Guanzhi Wang, De- An Huang, Osbert Bastani, Dinesh Jayaraman, Yuke Zhu, Linxi Fan, and Anima Anandkumar. Eureka: Human- level reward design via coding large language models. arXiv preprint arXiv:2310.12931 , 2023

    arXiv

  30. [38]

    Warp: A high-performance python frame- work for gpu simulation and graphics

    Miles Macklin. Warp: A high-performance python frame- work for gpu simulation and graphics. In NVIDIA GPU Technology Conference (GTC), 2022

    2022

  31. [39]

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

    Viktor Makoviychuk, Lukasz Wawrzyniak, Yunrong Guo, Michelle Lu, Kier Storey, Miles Macklin, David Hoeller, Nikita Rudin, Arthur Allshire, Ankur Handa, et al. Isaac gym: High performance gpu-based physics simulation for robot learning. arXiv preprint arXiv:2108.10470, 2021

    2021 arXiv

  32. [40]

    Learning robust perceptive locomotion for quadrupedal robots in the wild

    Takahiro Miki, Joonho Lee, Jemin Hwangbo, Lorenz Wellhausen, Vladlen Koltun, and Marco Hutter. Learning robust perceptive locomotion for quadrupedal robots in the wild. Science robotics, 7(62):eabk2822, 2022

    2022

  33. [41]

    Orbit: A unified simulation framework for interactive robot learning envi- ronments

    Mayank Mittal, Calvin Yu, Qinxi Yu, Jingzhou Liu, Nikita Rudin, David Hoeller, Jia Lin Yuan, Ritvik Singh, Yunrong Guo, Hammad Mazhar, et al. Orbit: A unified simulation framework for interactive robot learning envi- ronments. IEEE Robotics and Automation Letters , 8(6): 3740–...

    2023

  34. [42]

    Rusu, Joel Veness, Marc G

    V olodymyr Mnih, Koray Kavukcuoglu, David Silver, Andrei A. Rusu, Joel Veness, Marc G. Bellemare, Alex Graves, Martin Riedmiller, Andreas K. Fidjeland, Georg Ostrovski, Stig Petersen, Charles Beattie, Amir Sadik, Ioannis Antonoglou, Helen King, Dharshan Ku- maran, Daan Wierstr...

    2015 doi

  35. [43]

    MuJoCo XLA (MJX)

    MuJoCo XLA Authors. MuJoCo XLA (MJX). https: //mujoco.readthedocs.io/en/stable/mjx.html. Accessed: December 16, 2024

    2024

  36. [44]

    Scott Niekum and Isaac I.Y . Saito. ar track alvar, 2016. URL https://github.com/ros-perception/ar track alvar

    2016

  37. [45]

    Dexpbt: Scaling up dexterous manipulation for hand-arm systems with pop- ulation based training

    Aleksei Petrenko, Arthur Allshire, Gavriel State, Ankur Handa, and Viktor Makoviychuk. Dexpbt: Scaling up dexterous manipulation for hand-arm systems with pop- ulation based training. RSS, 2023

    2023

  38. [46]

    Asymmetric actor critic for image-based robot learning

    Lerrel Pinto, Marcin Andrychowicz, Peter Welinder, Wo- jciech Zaremba, and Pieter Abbeel. Asymmetric actor critic for image-based robot learning. RSS, 2018

    2018

  39. [47]

    Learning humanoid locomotion over challenging terrain

    Ilija Radosavovic, Sarthak Kamat, Trevor Darrell, and Jitendra Malik. Learning humanoid locomotion over challenging terrain. arXiv preprint arXiv:2410.03654 , 2024

    2024 arXiv

  40. [48]

    Prajit Ramachandran, Barret Zoph, and Quoc V . Le. Searching for activation functions, 2017. URL https: //arxiv.org/abs/1710.05941

    2017 arXiv

  41. [49]

    High-throughput batch rendering for embodied ai

    Luc Guy Rosenzweig, Brennan Shacklett, Warren Xia, and Kayvon Fatahalian. High-throughput batch rendering for embodied ai. 2024

    2024

  42. [50]

    Learning to walk in minutes using massively parallel deep reinforcement learning

    Nikita Rudin, David Hoeller, Philipp Reist, and Marco Hutter. Learning to walk in minutes using massively parallel deep reinforcement learning. In Conference on Robot Learning, pages 91–100. PMLR, 2022

    2022

  43. [51]

    Proximal policy optimization algorithms

    John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347 , 2017

    2017 arXiv

  44. [52]

    Humanoidbench: Simulated humanoid benchmark for whole-body locomo- tion and manipulation

    Carmelo Sferrazza, Dun-Ming Huang, Xingyu Lin, Youngwoon Lee, and Pieter Abbeel. Humanoidbench: Simulated humanoid benchmark for whole-body locomo- tion and manipulation. Robotics: Science and Systems (RSS), 2024

    2024

  45. [53]

    An ex- tensible, data-oriented architecture for high-performance, many-world simulation

    Brennan Shacklett, Luc Guy Rosenzweig, Zhiqiang Xie, Bidipta Sarkar, Andrew Szot, Erik Wijmans, Vladlen Koltun, Dhruv Batra, and Kayvon Fatahalian. An ex- tensible, data-oriented architecture for high-performance, many-world simulation. ACM Transactions on Graphics (TOG), 42(4...

    2023

  46. [54]

    An ex- tensible, data-oriented architecture for high-performance, many-world simulation

    Brennan Shacklett, Luc Guy Rosenzweig, Zhiqiang Xie, Bidipta Sarkar, Andrew Szot, Erik Wijmans, Vladlen Koltun, Dhruv Batra, and Kayvon Fatahalian. An ex- tensible, data-oriented architecture for high-performance, many-world simulation. ACM Trans. Graph. , 42(4), 2023

    2023

  47. [55]

    Learning free gait tran- sition for quadruped robots via phase-guided controller

    Yecheng Shao, Yongbin Jin, Xianwei Liu, Weiyan He, Hongtao Wang, and Wei Yang. Learning free gait tran- sition for quadruped robots via phase-guided controller. IEEE Robotics and Automation Letters , 7(2):1230–1237, 2021

    2021

  48. [56]

    Leap hand: Low-cost, efficient, and anthropomorphic hand for robot learning

    Kenneth Shaw, Ananye Agarwal, and Deepak Pathak. Leap hand: Low-cost, efficient, and anthropomorphic hand for robot learning. Robotics: Science and Systems (RSS), 2023

    2023

  49. [57]

    Dextrah-rgb: Visuomotor policies to grasp anything with dexterous hands

    Ritvik Singh, Arthur Allshire, Ankur Handa, Nathan Ratliff, and Karl Van Wyk. Dextrah-rgb: Visuomotor policies to grasp anything with dexterous hands. arXiv preprint arXiv:2412.01791, 2024

    2024 arXiv

  50. [58]

    Legged robots that keep on learning: Fine-tuning locomotion policies in the real world

    Laura Smith, J Chase Kew, Xue Bin Peng, Sehoon Ha, Jie Tan, and Sergey Levine. Legged robots that keep on learning: Fine-tuning locomotion policies in the real world. In 2022 International Conference on Robotics and Automation (ICRA), pages 1593–1599. IEEE, 2022

    2022

  51. [59]

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

    Jie Tan, Tingnan Zhang, Erwin Coumans, Atil Iscen, Yunfei Bai, Danijar Hafner, Steven Bohez, and Vincent Vanhoucke. Sim-to-real: Learning agile locomotion for quadruped robots. arXiv preprint arXiv:1804.10332 , 2018

    2018 arXiv

  52. [60]

    Maniskill3: Gpu parallelized robotics simulation and rendering for gener- alizable embodied ai

    Stone Tao, Fanbo Xiang, Arth Shukla, Yuzhe Qin, Xander Hinrichsen, Xiaodi Yuan, Chen Bao, Xinsong Lin, Yulin Liu, Tse-kai Chan, et al. Maniskill3: Gpu parallelized robotics simulation and rendering for gener- alizable embodied ai. arXiv preprint arXiv:2410.00425 , 2024

    arXiv

  53. [61]

    Deep- mind control suite

    Yuval Tassa, Yotam Doron, Alistair Muldal, Tom Erez, Yazhe Li, Diego de Las Casas, David Budden, Abbas Abdolmaleki, Josh Merel, Andrew Lefrancq, et al. Deep- mind control suite. arXiv preprint arXiv:1801.00690 , 2018

    2018 arXiv

  54. [62]

    Domain ran- domization for transferring deep neural networks from simulation to the real world

    Josh Tobin, Rachel Fong, Alex Ray, Jonas Schneider, Wojciech Zaremba, and Pieter Abbeel. Domain ran- domization for transferring deep neural networks from simulation to the real world. In IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages 23–30, 2017

    2017

  55. [63]

    Mujoco: A physics engine for model-based control

    Emanuel Todorov, Tom Erez, and Yuval Tassa. Mujoco: A physics engine for model-based control. In 2012 IEEE/RSJ international conference on intelligent robots and systems, pages 5026–5033. IEEE, 2012

    2012

  56. [64]

    Efficientzero v2: Mastering discrete and continuous control with limited data

    Shengjie Wang, Shaohuai Liu, Weirui Ye, Jiacheng You, and Yang Gao. Efficientzero v2: Mastering discrete and continuous control with limited data. arXiv preprint arXiv:2403.00564, 2024

    2024 arXiv

  57. [65]

    Bench- marking the performance and energy efficiency of ai accelerators for ai training

    Yuxin Wang, Qiang Wang, Shaohuai Shi, Xin He, Zhen- heng Tang, Kaiyong Zhao, and Xiaowen Chu. Bench- marking the performance and energy efficiency of ai accelerators for ai training. In 2020 20th IEEE/ACM International Symposium on Cluster, Cloud and Internet Computing (CCGRID...

    2020

  58. [66]

    Full-order sampling-based mpc for torque- level locomotion control via diffusion-style annealing

    Haoru Xue, Chaoyi Pan, Zeji Yi, Guannan Qu, and Guanya Shi. Full-order sampling-based mpc for torque- level locomotion control via diffusion-style annealing. arXiv preprint arXiv:2409.15610 , 2024

    2024 arXiv

  59. [67]

    Mastering visual continuous control: Improved data-augmented reinforcement learning

    Denis Yarats, Rob Fergus, Alessandro Lazaric, and Lerrel Pinto. Mastering visual continuous control: Improved data-augmented reinforcement learning. arXiv preprint arXiv:2107.09645, 2021

    2021 arXiv

  60. [68]

    MuJoCo Menagerie: A collection of high- quality simulation models for MuJoCo, 2022

    Kevin Zakka, Yuval Tassa, and MuJoCo Menagerie Con- tributors. MuJoCo Menagerie: A collection of high- quality simulation models for MuJoCo, 2022. URL http://github.com/google-deepmind/mujoco menagerie

    2022

  61. [69]

    Sim-to-real transfer in deep reinforcement learning for robotics: a survey

    Wenshuai Zhao, Jorge Pe ˜na Queralta, and Tomi Wester- lund. Sim-to-real transfer in deep reinforcement learning for robotics: a survey. In 2020 IEEE symposium series on computational intelligence (SSCI), pages 737–744. IEEE, 2020

    2020

  62. [70]

    Robot parkour learning

    Ziwen Zhuang, Zipeng Fu, Jianren Wang, Christo- pher Atkeson, Soeren Schwertfeger, Chelsea Finn, and Hang Zhao. Robot parkour learning. arXiv preprint arXiv:2309.05665, 2023. Appendix Table of Contents Appendix A: DM Control Suite 12 A.1 Environments . . . . . . . . . . . . . ...

    2023 arXiv

  63. [72]

    Train for 100 M timesteps with a command range of {1.5, 0.8, 1.2}

  64. [73]

    Finetune for 50 M timesteps with a command range of {1.5, 0.8, 2π}

  65. [74]

    Finetune on rough terrain for 100 M timesteps. B.25b. Getup policy:

  66. [75]

    Train with a power termination cutoff of 400 W

  67. [76]

    Finetune with a joint velocity cost. B.25c. Handstand and footstand policies:

  68. [77]

    Finetune with a joint acceleration and energy cost

  69. [78]

    injections

    Progressively reduce the power termination budget from 400 W to 200 W. Finally, all policies are trained on flat terrain for 200 M timesteps, then finetuned on rough terrain for 100 M timesteps. The rough terrain is modeled as a heightfield generated from Perlin noise. B.3. RL...

  70. [2024]

    doi: 10.1109/TASE.2024.3410297

    2024

Pith tools

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