Pith. sign in

REVIEW 5 major objections 4 minor 45 cited by

ASAP: Aligning Simulation and Real-World Physics for Learning Agile Humanoid Whole-Body Skills

T0 review · 5 major / 4 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read A residual action model learned from real-world rollouts and frozen into the simulator reduces motion-tracking error for agile humanoid skills.

desk verdict A solid, well-engineered application of residual action learning to whole-body humanoid sim-to-real with real code and a real robot demo, but the real-world evidence is thinner than the abstract suggests and the headline '52.7%' is not supported by the real-world table. read the letter →

arxiv 2502.01143 v3 pith:D5XAGIUP submitted 2025-02-03 cs.RO cs.AIcs.LGcs.SYeess.SY

Tairan He , Jiawei Gao , Wenli Xiao , Yuanhang Zhang , Zi Wang , Jiashun Wang , Zhengyi Luo , Guanqi He
show 10 more authors
Nikhil Sobanbab Chaoyi Pan Zeji Yi Guannan Qu Kris Kitani Jessica Hodgins Linxi "Jim" Fan Yuke Zhu Changliu Liu Guanya Shi
This is my paper · ORCID
classification cs.ROcs.AIcs.LGcs.SYeess.SY
keywords sim-to-realtransferhumanoidwhole-bodycontrolresidualactionmodelreinforcementlearningmotiontrackingagileskillsdynamicsmismatch
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 tries to establish that the gap between simulated and real physics for agile humanoid whole-body skills can be closed by learning a small residual action correction from real-world rollouts and freezing it into the simulator before fine-tuning the control policy. If true, this gives a practical recipe that avoids laborious system identification and avoids the conservative behavior that domain randomization can induce. The recipe is: pre-train a motion-tracking policy in simulation from human-video-derived reference motions, collect roughly one hundred real motion clips, train a delta action model that compensates for the state discrepancy when those clips are replayed in simulation, fine-tune the policy in the corrected simulator, then deploy the fine-tuned policy alone. The authors demonstrate the recipe across two sim-to-sim transfers and on a real 23-degree-of-freedom humanoid, with lower tracking error than baselines and with a 4-degree-of-freedom ankle-only correction sufficient for the real motions tested.

What carries the argument

The load-bearing object is the RL-trained delta action model $\pi_\Delta(s_t,a_t)$, a residual policy that maps the current state and commanded action to a corrective action. It is trained by initializing the simulator at a recorded real-world state $s^r_t$ and rewarding agreement between the simulated next state and the recorded real next state $s^r_{t+1}$, with an action-norm penalty to keep corrections small. Once trained, the model is frozen and wrapped into the simulator transition $s_{t+1}=f_{\mathrm{sim}}(s_t,a_t+\pi_\Delta(s_t,a_t))$, so that policy fine-tuning experiences the shifted dynamics; at deployment the delta model is removed and only the fine-tuned policy runs.

What would settle it

Apply the same training protocol but freeze a full 23-joint delta action model trained with comparable real data (or estimate per-joint errors from the 100 clips) and compare real-world global tracking error to the 4-DoF version; if the full model's error is materially lower, the ankle-only restriction is the wrong load-bearing choice.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that the sim-to-real dynamics gap for agile whole-body humanoid control can be treated as a correction to actions rather than to parameters or states. The authors train a delta action model $\pi_\Delta$ from real-world rollouts, freeze it into the simulator as $s_{t+1}=f_{\mathrm{sim}}(s_t,a_t+\pi_\Delta(s_t,a_t))$, and fine-tune the pre-trained motion-tracking policy inside that corrected simulator. They report that this reduces tracking error against system identification, domain randomization, and learned-residual-dynamics baselines in two sim-to-sim transfers and on a real 23-degree-of-freedom humanoid, where a 4-degree-of-freedom ankle-only correction suffices for the motions tested. The measured gains include a drop in global body-position error from about 159 mm to 112 mm on an out-of-distribution real-world motion, and the method achieves 100 percent success on hard sim-to-sim tasks where baselines fail. If true, the useful story is that a small, structured residual in action space can absorb the physics mismatch that system identification and domain randomization cannot.

Load-bearing premise

The claim rests on the assumption that correcting the robot's four ankle degrees of freedom is enough to capture the dominant mismatch between simulation and real physics; if knee and hip discrepancies are substantial, the real-world benefit of the method shrinks.

Editorial extensions

If this is right

  • A policy fine-tuned in a simulator that contains the frozen delta action model tracks reference motions with lower global body-position and joint-angle errors in the target environment, including a real-world out-of-distribution motion whose global error falls from about 159 mm to 112 mm.
  • The method keeps 100 percent task success on hard sim-to-sim transfers where system identification and delta-dynamics baselines drop to 60 to 83 percent, so the correction mechanism, not just extra training, is what preserves agility.
  • RL fine-tuning with the delta model beats training-free fixes such as fixed-point iteration and gradient search, showing the compensation must be optimized closed-loop rather than solved one step at a time.
  • The learned corrections are per-joint and asymmetric (ankle and knee deltas are largest, upper-body deltas small), so uniform random action noise cannot reproduce the effect.
  • Sim-to-sim transfer between different physics engines can serve as a fast proxy for real-world delta-action alignment before any hardware rollout is collected.

Reading between the lines

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

  • A natural extension the paper does not pursue: the same delta-action idea could run online, updating $\pi_\Delta$ on each new hardware unit or as motors wear, turning a one-time calibration into continuous self-alignment.
  • The dataset-size saturation in the paper (4300 versus 43000 samples changes closed-loop error by only about 0.65 percent) suggests that active collection targeting high-error joints and motion phases could make real-world data use far more efficient.
  • The paper's own per-joint magnitude plot shows knee and hip discrepancies inside the full 23-DoF model, so a low-rank correction over ankle, knee, and hip trained from the same 100 clips is a direct test of whether the 4-DoF restriction costs agility on hard jumps.
  • Because the delta model only changes the transition function inside the simulator, it should compose with stronger pre-training (for example, better domain randomization or richer reference data) rather than competing with it.
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

5 major / 4 minor

Summary. ASAP proposes a two-stage framework for closing the sim-to-real dynamics gap in agile humanoid whole-body motion tracking. Stage one pre-trains phase-conditioned motion-tracking policies in IsaacGym using retargeted human motion; stage two collects real-world rollouts under the pre-trained policy, trains a delta (residual) action model with RL so that the simulator reproduces real-world next states, freezes this model into the simulator as s_{t+1}=f_sim(s_t, a_t+pi_delta(s_t,a_t)), and fine-tunes the policy; deployment then uses the fine-tuned policy without the delta model. The paper evaluates on IsaacGym-to-IsaacSim, IsaacGym-to-Genesis, and IsaacGym-to-Unitree G1, reporting lower tracking errors than Vanilla, SysID, and DeltaDynamics baselines, with open-loop replay metrics, closed-loop policy metrics, and a small real-world table.

Significance. If the central claim holds, residual action alignment is an attractive alternative to SysID and DR: it avoids hand-picked parameter spaces, preserves agility, and is trainable from policy rollouts. The paper's strengths include an open-source multi-simulator codebase, a systematic comparison against SysID and learned-dynamics baselines, and ablations (dataset size, horizon, action norm weight) that show the delta action model improves out-of-distribution generalization. The real-world G1 demonstrations of kicks, jumps, and balance are impressive and constitute a demanding testbed. However, the evidence for the central real-world claim currently rests on point estimates from a reduced 4-DoF ankle model, and some empirical statements in the paper are contradicted by its own tables.

major comments (5)
  1. [Section IV-A, Table III] The sentence "Quantitative results in Table III demonstrate that ASAP consistently outperforms the OpenLoop baseline across all replayed motion lengths" is not supported by Table III: at the 0.25 s replay length in IsaacSim, ASAP has Eg-mpjpe=19.9 mm and Empjpe=15.6 mm versus OpenLoop's 19.5 mm and 15.1 mm, so it is slightly worse on both metrics; at 0.5 s in IsaacSim, DeltaDynamics has lower Empjpe (16.4) than ASAP (19.2). Please either restrict the claim to the longer horizons and specify the primary metric, or add significance or paired comparisons.
  2. [Section IV-B, Table IV] The claim that ASAP "consistently outperforms baselines across all difficulty levels" and achieves the "lowest Eg-mpjpe and Empjpe" is contradicted by Table IV on root-relative MPJPE: in IsaacSim Hard, Vanilla Empjpe=51.6 versus ASAP=56.5; in Genesis Easy, Vanilla Empjpe=70.1 versus ASAP=73.5; and in IsaacSim Medium, Vanilla Empjpe=49.2 versus ASAP=49.3. Even on Eg-mpjpe, several Easy and Medium differences are only 1-7 mm, which is within or near one standard error. Please clarify whether Eg-mpjpe is the sole primary outcome and provide error bars or statistical tests for the closed-loop comparisons.
  3. [Section IV-C, Table V] The real-world evaluation is the direct evidence for Q3, yet Table V reports single point estimates per metric with no variance, confidence intervals, or significance tests, even though Section IV-C states that the tracking policy was executed 30 times per task. With n=30, paired mean differences and standard errors should be reported; without them, reductions of 7.8-29.6% cannot be distinguished from run-to-run variability.
  4. [Section IV-C, Figure 13] The decision to train a 4-DoF ankle-only delta action model rests on the claim that the G1 ankle linkage dominates the sim-to-real gap, but no real-world measurement is provided to support that dominance. Figure 13, for the IsaacGym-to-IsaacSim case, shows pronounced corrections at the knee as well as the ankle, and the paper's own limitation section notes that the full 23-DoF model is data-hungry. If the ankle-only model misses a substantial share of the mismatch, the fine-tuned policy is adapted to only a partially aligned simulator. Please add per-joint delta action statistics from real-world data or otherwise validate the 4-DoF restriction before using it as the basis for the real-world claim.
  5. [Conclusion] The conclusion states reductions "up to 52.7% in sim-to-real tasks", but the maximum reduction in Table V is about 29.6% (LeBron Eg-mpjpe: 159 to 112). The 52.7% figure appears to correspond to a sim-to-sim open-loop comparison in Table III (for example, IsaacSim 1.0 s: 80.8 to 37.9, which is a 53.1% reduction). Please correct the conclusion or explicitly attribute this number to the appropriate experiment.
minor comments (4)
  1. [Section IV-C] The sentence "the minimum required to train the full 23-DoF delta action model in simulation, as discussed in Section III-B" appears in Section IV-C, but Section III-B does not discuss data requirements; this belongs to the dataset-size analysis in Section V-A or the limitations section. Please fix the cross-reference.
  2. [Figure 13] Please add a color scale and numeric labels to Figure 13 so that the claimed ankle and knee dominance can be assessed quantitatively.
  3. [Abstract and Section IV-B] The abstract and Section IV-B use "significantly" without reporting significance tests; please either add statistical tests or use more neutral phrasing such as "consistently reduces" where appropriate.
  4. [Section VIII-D] Equation (1) in Section VIII-D states f_real(s,pi(s)) = f_sim(s,pi(s)+pi_delta(s,pi(s))), but this is an assumed one-step consistency condition, not a derived identity; labeling it explicitly as an assumption would clarify the derivation.

Circularity Check

1 steps flagged · score 4.0 of 10

Open-loop dynamics-matching benchmark reports training-fit error as generalization; closed-loop and OOD real-world claims remain independent.

  1. fitted input called prediction [Section IV-A (Table III) and Section III-B (Training Delta Action Model)]
    "Open-loop evaluation measures how accurately a method can reproduce testing-environment trajectories in the training environment. This is achieved by rolling out the same trajectory executed in the testing environment and assessing tracking discrepancies using key metrics such as MPJPE. ... A reward signal is computed to minimize the discrepancy between the simulated state st+1 and the recorded real-world state sr_t+1. ... ASAP, however, demonstrates superior generalization by learning residual policies that effectively bridge the dynamics gap."

    The delta action model is trained by RL to output corrections such that f_sim(st, ar_t + pi_delta(st, ar_t)) matches the recorded next state sr_t+1 on the recorded trajectories. The open-loop benchmark then replays 'the same trajectory executed in the testing environment' — under the sim-to-sim proxy, these are the very recorded trajectories used to train pi_delta — and scores ASAP by tracking discrepancy against those recorded states. The ASAP row in Table III is therefore a training-fit error on the exact objective the delta model was optimized for, not an independent prediction. Describing this result as 'superior generalization' overstates what the benchmark can show.

full rationale

The only exhibited circular step is in the open-loop dynamics-matching benchmark: the delta action model's training reward is next-state matching on the recorded trajectories, and the open-loop metric replays those same trajectories and reports the mismatch, so the ASAP row of Table III is a fit rather than a held-out prediction, and the 'superior generalization' sentence attached to it is not supported by that table. All other claimed results are not circular: the delta model is an empirical residual fit to recorded dynamics, the fine-tuned policy is evaluated on real tracking error which is not the delta model's objective, the in-distribution kick result is supplemented by an out-of-distribution LeBron 'Silencer' evaluation, and no uniqueness theorem or load-bearing self-citation is invoked. The retargeting pipeline cites the authors' earlier H2O work, but that is a methodological tool rather than a self-referential justification of the delta-action claim. The conclusion's 'up to 52.7%' figure is inconsistent with Table V (max about 29.6%), but that is a correctness or consistency concern, not circularity. Overall the central derivation is self-contained; the score reflects the one supporting benchmark that reduces to its training objective.

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

The framework introduces no new physical entities. Its central ingredient is a learned residual action model, which is an empirical fit rather than a postulated mechanism. The key contingent choices are the action norm weight, training horizon, reward weights, domain randomization ranges, termination curriculum, and the 4-DoF ankle reduction, all of which influence the reported results.

free parameters (5)
  • Action norm weight = 0.1
    Selected by ablation in Figure 10(c); controls the trade-off between dynamics alignment and minimal correction magnitude in the delta action RL objective (Table II).
  • Delta action training horizon = 1.0s
    Selected by ablation in Figure 10(b); open-loop error improves with longer horizons, but closed-loop fine-tuning is best at 1.0s.
  • Reward weights (pretraining and delta action) = Tables I and II
    Hand-specified weights for penalty, regularization, and task rewards; no sensitivity analysis is provided for most weights.
  • Domain randomization ranges = Table VI
    Friction U(0.2,1.1), P gain U(0.925,1.05), control delay 20-40ms, push perturbation 0.5 m/s; these shape the pretrained policy and are not derived from data.
  • Termination curriculum thresholds = 1.5m to 0.3m
    Hand-set tolerance schedule for tracking termination in pretraining (Section II-B, b); affects which motions the policy can learn.
assumptions (6)
  • domain assumption The sim-to-real dynamics gap can be represented as an additive correction on actions: f_sim(s, a + pi_delta(s,a)) approximates f_real(s,a).
    Core modeling assumption in Section III-C; the entire ASAP framework rests on this residual action parameterization.
  • domain assumption A 4-DoF ankle delta action model is sufficient for real-world whole-body alignment.
    Section IV-C adopts 4-DoF ankle correction due to data constraints; this limits the generality of the whole-body claim.
  • domain assumption Motion capture and onboard sensors provide accurate real-world states.
    Section III-A relies on MoCap and onboard sensing to record trajectories used to train the delta action model.
  • domain assumption The time-phase variable phi is a sufficient goal for single-motion tracking.
    Section II-B builds the goal condition on phi, citing DeepMimic [67]; this design choice is not re-validated here.
  • domain assumption 100 real-world clips are sufficient to train an effective 4-DoF delta action model.
    Section IV-C states 100 clips 'prove sufficient' with no systematic scaling study in the real domain.
  • standard math PPO converges to a policy that minimizes the delta action learning objective.
    Standard RL assumption; the paper uses PPO without formal convergence guarantees.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ASAP: Aligning Simulation and Real-World Physics for Learning Agile Humanoid Whole-Body Skills." pith.science (2026). https://pith.science/paper/D5XAGIUP

@misc{pith2026250201143,
  author       = {Pith},
  title        = {Pith review of: ASAP: Aligning Simulation and Real-World Physics for Learning Agile Humanoid Whole-Body Skills},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/D5XAGIUP}},
  note         = {Machine review of arXiv:2502.01143}
}
read the original abstract

Humanoid robots hold the potential for unparalleled versatility in performing human-like, whole-body skills. However, achieving agile and coordinated whole-body motions remains a significant challenge due to the dynamics mismatch between simulation and the real world. Existing approaches, such as system identification (SysID) and domain randomization (DR) methods, often rely on labor-intensive parameter tuning or result in overly conservative policies that sacrifice agility. In this paper, we present ASAP (Aligning Simulation and Real-World Physics), a two-stage framework designed to tackle the dynamics mismatch and enable agile humanoid whole-body skills. In the first stage, we pre-train motion tracking policies in simulation using retargeted human motion data. In the second stage, we deploy the policies in the real world and collect real-world data to train a delta (residual) action model that compensates for the dynamics mismatch. Then, ASAP fine-tunes pre-trained policies with the delta action model integrated into the simulator to align effectively with real-world dynamics. We evaluate ASAP across three transfer scenarios: IsaacGym to IsaacSim, IsaacGym to Genesis, and IsaacGym to the real-world Unitree G1 humanoid robot. Our approach significantly improves agility and whole-body coordination across various dynamic motions, reducing tracking error compared to SysID, DR, and delta dynamics learning baselines. ASAP enables highly agile motions that were previously difficult to achieve, demonstrating the potential of delta action learning in bridging simulation and real-world dynamics. These results suggest a promising sim-to-real direction for developing more expressive and agile humanoids.

Figures

Figures reproduced from arXiv: 2502.01143 by the authors.

Figure 1
Figure 1. The humanoid robot (Unitree G1) demonstrates diverse agile whole-body skills, showcasing the control policies’ agility: (a) Cristiano Ronaldo’s [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of ASAP. (a) Motion Tracking Pre-training and Real Trajectory Collection: With the humanoid motions retargeted from human videos, we pre-train multiple motion tracking policies to roll out real-world trajectories. (b) Delta Action Model Training: Based on the real-world rollout data, we train the delta action model by minimizing the discrepancy between simulation state st and real-world state s r t . (c) Po… view at source ↗
Figure 3
Figure 3. Retargeting Human Video Motions to Robot Motions: (a) Human motions are captured from video. (b) Using TRAM [ [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Baselines of ASAP. (a) Model-free RL training. (b) System ID from real to sim using real-world data. (c) Learning delta dynamics model using real-world data. (d) Our proposed method, learning delta action model using real-world data. physics. We leverage an RL-based de…
Figure 5
Figure 5. Figure 5: Replaying IsaacSim State-Action trajecories in IsaacGym. The upper four panels visualize the Unitree G1 humanoid executing a soccer-shooting [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Visual comparisons of motion imitation results across different difficulty levels (Easy, Medium, Hard) for various tasks including Jump Forward, [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Visualization of G1 motion tracking before and after [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: Visualization of LeBron James’ “Silencer” motion on the G1 robot before (upper figure enclosed in [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]
Figure 9
Figure 9. Figure 9: We deploy the pretrained policy of a forward jump motion tracking [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 10
Figure 10. Figure 10: Analysis of dataset size, training horizon, and action norm on the performance of [PITH_FULL_IMAGE:figures/full_fig_p010_10.png]
Figure 12
Figure 12. Figure 12: MPJPE vs. Noise Level for policies fine-tuned with random action [PITH_FULL_IMAGE:figures/full_fig_p011_12.png]
Figure 13
Figure 13. Figure 13: Visualization of IsaacGym-to-IsaacSim π∆ output magnitude. We compute the average absolute value of each joint over the 4300-episode dataset. Larger red dots indicate higher values. The results suggest that lower￾body motors exhibit a larger discrepancy compared to up…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 45 Pith papers

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

  1. Light-Loco-Parkour: Versatile Perceptive Whole-Body Locomotion via Multi-Skill Distillation

    cs.RO 2026-08 conditional novelty 7.0 of 10

    A single neural-network policy, trained in simulation, makes a humanoid climb, vault, and traverse uneven terrain from onboard depth and a velocity command, with no skill labels or runtime motion graphs.

  2. Handroid: Bridging Dexterous Hand and Humanoid

    cs.RO 2026-07 conditional novelty 7.0 of 10

    A single 27-DoF body doubles as an anthropomorphic dexterous hand and a 0.33 m desktop humanoid, with a unified control stack for manipulation, locomotion, and embodiment switching.

  3. ARDY: Autoregressive Diffusion with Hybrid Representation for Interactive Human Motion Generation

    cs.GR 2026-07 accept novelty 7.0 of 10

    An autoregressive diffusion model with a hybrid explicit-root/latent-body representation generates real-time, controllable 3D human motion from text and spatial constraints.

  4. MOVE: Motion-Guided Few-Shot Video Object Segmentation

    cs.CV 2025-07 conditional novelty 7.0 of 10

    MOVE provides a new motion-guided few-shot video object segmentation benchmark, and the proposed DMA baseline outperforms six existing methods across all settings.

  5. Sampling-Based System Identification with Active Exploration for Legged Robot Sim2Real Learning

    cs.RO 2025-05 conditional novelty 7.0 of 10

    SPI-Active identifies legged-robot physical parameters via massive parallel sampling and uses Fisher-information-optimal command sequences to collect informative real-world data, improving sim-to-real transfer on quad...

  6. 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.

  7. $\omega$-0: A Latent Predictive World Action Model for Concurrent Humanoid Loco-Manipulation

    cs.RO 2026-08 conditional novelty 6.0 of 10

    A single whole-body model with latent future prediction outperforms prior robot policies on 11 real-world humanoid household loco-manipulation tasks.

  8. Weights or Skills? A Survey of Robot-Learning Techniques: from Action-Predicting Weights to Robots that Write their Own Skills

    cs.RO 2026-08 conditional novelty 6.0 of 10

    A taxonomy of robot learning on a weights-versus-skills axis, with a five-rung self-improvement ladder whose top cell (feedback plus memory plus search) holds only a few recent systems.

  9. GenTrack: Physical Alignment for Robot-Native Motion Generation and Zero-Shot Humanoid Tracking

    cs.RO 2026-08 conditional novelty 6.0 of 10

    Online co-training of a text-to-motion generator and a humanoid tracker on simulated G1 improves generator executability and zero-shot tracker coverage beyond static replay or one-way filtering.

  10. Bicycle Acrobatics with Reinforcement Learning

    cs.RO 2026-08 conditional novelty 6.0 of 10

    A reinforcement-learning pipeline trained on a custom bicycle robot, then orchestrated by a state machine, performs repeated acrobatic stunts including jumps, flips, wheelies, and kip-ups in hardware.

  11. Scaling Behavior Foundation Model for Humanoid Robots

    cs.RO 2026-07 conditional novelty 6.0 of 10

    A scaling recipe for humanoid behavior foundation models—global-frame motion tracking, on-policy data quantity plus reference diversity, and a transformer with hyperspherical latents—cuts global tracking error by roug...

  12. Behavior Foundations for Quadruped Robots: ABot-C0 Technical Report

    cs.RO 2026-07 conditional novelty 6.0 of 10

    A multi-source 16,074-clip quadruped motion library plus a flow-matching generalist tracker shows empirical data scaling and zero-shot unseen tracking, integrated with all-terrain locomotion and real-robot deployment.

  13. Closing the Reality Gap: Zero-Shot Sim-to-Real Deployment for Dexterous Force-Based Grasping and Manipulation

    cs.RO 2026-07 conditional novelty 6.0 of 10

    Zero-shot sim-to-real RL policies on a five-finger hand achieve commandable grasp-force tracking and in-hand reorientation using dense tactile simulation, current-to-torque calibration, and actuator randomization.

  14. VOFA: Visual Object Goal Pushing with Force-Adaptive Control for Humanoids

    cs.RO 2026-05 unverdicted novelty 6.0 of 10

    VOFA combines a high-level visuomotor policy with a low-level force-adaptive controller to let humanoids push objects up to 17 kg to arbitrary goals using only noisy onboard vision, achieving over 80% real-world success.

  15. Kinodynamic Motion Retargeting for Humanoid Locomotion via Multi-Contact Whole-Body Trajectory Optimization

    cs.RO 2026-03 conditional novelty 6.0 of 10

    A physics-aware motion-retargeting pipeline that uses ground-reaction-force-derived heel-toe contacts produces dynamically feasible humanoid references and improves downstream imitation learning.

  16. Humanoid Everyday: A Comprehensive Robotic Dataset for Open-World Humanoid Manipulation

    cs.RO 2025-10 conditional novelty 6.0 of 10

    A 10,300-demonstration, 260-task multimodal humanoid manipulation dataset with baseline policy evaluations and a cloud evaluation platform.

  17. A Scalable Whole-body Motion Transfer via Implicit Kinodynamic Motion Retargeting

    cs.RO 2025-09 conditional novelty 6.0 of 10

    A neural retargeting pipeline maps human motion to humanoid robot motion at 5000+ frames per second using a shared latent space and physics-based fine-tuning, filtering noise and producing physically feasible trajectories.

  18. Towards bridging the gap: Systematic sim-to-real transfer for diverse legged robots

    cs.RO 2025-09 conditional novelty 6.0 of 10

    PACE fits a compact set of actuator parameters from brief in-air data and trains energy-aware locomotion policies that transfer zero-shot to real quadrupeds without dynamics randomization.

  19. In-between Motion Generation Based Multi-Style Quadruped Robot Locomotion

    cs.RO 2025-07 conditional novelty 6.0 of 10

    A CVAE-based in-between motion generator creates multi-style quadruped gaits from sparse motion data, and the trained controller runs gallop, tripod, trotting, and pacing on a real robot.

  20. Robot Drummer: Learning Rhythmic Skills for Humanoid Drumming

    cs.RO 2025-07 conditional novelty 6.0 of 10

    A simulated Unitree G1 humanoid learns to drum dozens of popular songs from MIDI with high F1 scores using a Rhythmic Contact Chain and temporal decomposition.

  21. Learning Motion Skills with Adaptive Assistive Curriculum Force in Humanoid Robots

    cs.RO 2025-06 conditional novelty 6.0 of 10

    A2CF uses an adaptive assistive-force agent to guide humanoid robots through training, yielding faster convergence and robust policies that work without the external force.

  22. GMT: General Motion Tracking for Humanoid Whole-Body Control

    cs.RO 2025-06 conditional novelty 6.0 of 10

    GMT trains a single unified humanoid policy using adaptive sampling and mixture-of-experts, achieving lower tracking errors than a re-implemented ExBody2 across diverse whole-body motions.

  23. KungfuBot: Physics-Based Humanoid Whole-Body Control for Learning Highly-Dynamic Skills

    cs.RO 2025-06 conditional novelty 6.0 of 10

    A robot control method that adaptively tightens motion-tracking reward tolerances achieves lower tracking errors on dynamic skills and transfers zero-shot to a real humanoid.

  24. From Experts to a Generalist: Toward General Whole-Body Control for Humanoid Robots

    cs.RO 2025-06 conditional novelty 6.0 of 10

    BumbleBee, an expert-to-generalist pipeline using autoencoder-based motion clustering and per-cluster delta action models, reports state-of-the-art whole-body control on a Unitree G1 humanoid, with success rates of 89...

  25. RL from Physical Feedback: Aligning Large Motion Models with Humanoid Control

    cs.RO 2025-06 reject novelty 6.0 of 10

    RLPF uses reinforcement learning with a physics-simulator tracking reward and an alignment verification module to fine-tune a large text-to-motion model for physically feasible humanoid motions.

  26. SkillBlender: Towards Versatile Humanoid Whole-Body Loco-Manipulation via Skill Blending

    cs.RO 2025-06 conditional novelty 6.0 of 10

    SkillBlender pretrains reusable goal-conditioned skills and blends them with softmax per-joint weights to solve simulated humanoid loco-manipulation tasks with one or two reward terms.

  27. MoRE: Mixture of Residual Experts for Humanoid Lifelike Gaits Learning on Complex Terrains

    cs.RO 2025-06 conditional novelty 6.0 of 10

    A two-stage reinforcement learning pipeline with a mixture of latent residual experts gives a Unitree G1 humanoid multiple commanded human-like gaits over complex terrains.

  28. Hold My Beer: Learning Gentle Humanoid Locomotion and End-Effector Stabilization Control

    cs.RO 2025-05 conditional novelty 6.0 of 10

    A slow-fast two-agent reinforcement learning architecture with separate upper- and lower-body policies reduces end-effector shaking during humanoid locomotion.

  29. Motion Priors Reimagined: Adapting Flat-Terrain Skills for Complex Quadruped Mobility

    cs.RO 2025-05 conditional novelty 6.0 of 10

    A two-level reinforcement learning system that first learns animal-like gaits from flat-ground motion data, then learns small joint corrections that let a quadruped robot traverse rough terrain and navigate to goals.

  30. What Matters for Batch Online Reinforcement Learning in Robotics?

    cs.RO 2025-05 conditional novelty 6.0 of 10

    A recipe combining an imitation-trained diffusion policy, an IQL-learned Q-function, and implicit Q-guided action selection enables robots to self-improve from autonomous rollouts, outperforming imitation and filtered...

  31. HuB: Learning Extreme Humanoid Balance

    cs.RO 2025-05 conditional novelty 6.0 of 10

    HuB combines reference motion refinement, balance shaping rewards, and robustness training to enable a G1 humanoid to hold extreme single-leg poses that prior tracking methods fail to maintain.

  32. ComplexMimic: Human-Scene Interaction Imitation in Complex 3D Environments

    cs.CV 2026-07 unverdicted novelty 5.5 of 10

    Dual-expert RL plus difficulty-aware multi-teacher distillation improves physics-based human–scene interaction imitation under complex 3D geometry versus prior single-policy baselines.

  33. Tired Actor: Fatigue-Informed Character Control

    cs.RO 2026-08 conditional novelty 5.0 of 10

    Injecting a muscle-fatigue model into a general physics-based character controller preserves motion imitation accuracy while producing tired, more human-like behaviors such as shorter steps, corner cutting, and fall c...

  34. Shooting for Contact: Contact-Implicit Multiple Shooting for Dynamic Motion Retargeting

    cs.RO 2026-08 conditional novelty 5.0 of 10

    A differentiable MuJoCo simulator is embedded in a multiple-shooting optimizer to generate dynamically feasible, contact-consistent reference motions that accelerate motion-imitation RL training and transfer zero-shot...

  35. A Single Diffusion-Policy Controller for Multi-Task Block Pushing with Zero-Shot Sim-to-Real Transfer

    cs.RO 2026-07 conditional novelty 5.0 of 10

    One diffusion policy trained via energy-guided RL solves multi-shape block pushing without demos and transfers zero-shot to real robots under varied conditions.

  36. ZeroWBC: Learning Natural Whole-Body Humanoid Interaction from Human Egocentric Data

    cs.RO 2026-03 conditional novelty 5.0 of 10

    An open-loop generation-then-tracking system maps one egocentric image plus language into Unitree G1 whole-body interactions using only human egocentric motion data.

  37. RobotDancing: Residual-Action Reinforcement Learning Enables Robust Long-Horizon Humanoid Motion Tracking

    cs.RO 2025-09 conditional novelty 5.0 of 10

    Residual-action reinforcement learning, with selective corrections on hip and knee pitch joints, enables zero-shot long-horizon dance tracking on real humanoid robots.

  38. HERMES: Human-to-Robot Embodied Learning from Multi-Source Motion Data for Mobile Dexterous Manipulation

    cs.RO 2025-08 conditional novelty 5.0 of 10

    HERMES converts a single human motion demonstration into a deployable mobile bimanual dexterous manipulation policy, using RL, depth-image distillation, and closed-loop PnP pose refinement.

  39. Quantifying and Visualizing Sim-to-Real Gaps: Physics-Guided Regularization for Reproducibility

    cs.RO 2025-07 reject novelty 5.0 of 10

    A gain-regularized, parameter-conditioned RNN balances a low-cost 110:1 gearbox robot with matching simulated and real settling times, while naive domain randomization oscillates.

  40. Hierarchical Vision-Language Planning for Multi-Step Humanoid Manipulation

    cs.RO 2025-06 conditional novelty 5.0 of 10

    A three-layer hierarchical system using a VLM planner and VLM skill monitor with imitation-learned skills and an RL tracking policy achieved 73% success on a real humanoid pick-and-place task.

  41. SLAC: Safe and Efficient Real-Robot Reinforcement Learning via Unsupervised Simulation Pre-Training

    cs.RO 2025-06 conditional novelty 5.0 of 10

    SLAC learns a latent action space in a low-fidelity simulator and uses it for real-world reinforcement learning, solving whole-body mobile manipulation tasks in under an hour without demonstrations.

  42. SMAP: Self-supervised Motion Adaptation for Physically Plausible Humanoid Whole-body Control

    cs.RO 2025-05 conditional novelty 5.0 of 10

    SMAP uses a vector-quantized periodic autoencoder to adapt human motion into physically plausible humanoid motion, then distills an RL teacher policy into a student policy for whole-body control.

  43. APEX: Action Priors Enable Efficient Exploration for Robust Motion Tracking on Legged Robots

    cs.RO 2025-05 conditional novelty 5.0 of 10

    APEX trains gait-tracking policies with decaying action priors and separate style and task critics, achieving reference-free deployment, faster convergence, and reward-robustness over DeepMimic.

  44. Theory Foundation of Physics-Enhanced Residual Learning

    cs.LG 2025-08 reject novelty 4.0 of 10

    A set of conditional bounds shows PERL's advantages follow from assumed smaller Lipschitz constant and loss ceiling, without proving those assumptions or connecting them correctly to neural network training.

  45. A Survey: Learning Embodied Intelligence from Physical Simulators and World Models

    cs.RO 2025-07 conditional novelty 4.0 of 10

    Embodied intelligence learning is reviewed through the complementary lenses of physical simulators and world models, with a proposed IR-L0 to IR-L4 robot capability taxonomy.

Reference graph

Works this paper leans on

113 extracted references · 34 canonical work pages · cited by 45 Pith papers

  1. [1]

    Residual reinforce- ment learning from demonstrations

    Minttu Alakuijala, Gabriel Dulac-Arnold, Julien Mairal, Jean Ponce, and Cordelia Schmid. Residual reinforce- ment learning from demonstrations. arXiv preprint arXiv:2106.08050, 2021

    arXiv 2021

  2. [2]

    Estimation of inertial parameters of rigid body links of manipulators

    Chae H An, Christopher G Atkeson, and John M Holler- bach. Estimation of inertial parameters of rigid body links of manipulators. In 1985 24th IEEE Conference on Decision and Control, pages 990–995. IEEE, 1985

    1985

  3. [3]

    From imitation to refinement–residual rl for precise assembly

    Lars Ankile, Anthony Simeonov, Idan Shenfeld, Mar- cel Torne, and Pulkit Agrawal. From imitation to refinement–residual rl for precise assembly. arXiv preprint arXiv:2407.16677, 2024

    arXiv 2024

  4. [4]

    A bayesian treatment of real-to-sim for deformable object manip- ulation

    Rika Antonova, Jingyun Yang, Priya Sundaresan, Dieter Fox, Fabio Ramos, and Jeannette Bohg. A bayesian treatment of real-to-sim for deformable object manip- ulation. IEEE Robotics and Automation Letters, 7(3): 5819–5826, 2022

    2022

  5. [5]

    System identi- fication—a survey

    Karl Johan ˚Astr¨om and Peter Eykhoff. System identi- fication—a survey. Automatica, 7(2):123–162, 1971

    1971

  6. [6]

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

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

  7. [7]

    Learning and deploying robust locomotion policies with minimal dynamics ran- domization

    Luigi Campanaro, Siddhant Gangapurwala, Wolfgang Merkt, and Ioannis Havoutis. Learning and deploying robust locomotion policies with minimal dynamics ran- domization. In 6th Annual Learning for Dynamics & Control Conference, pages 578–590. PMLR, 2024

    2024

  8. [8]

    Residual robot learning for object-centric probabilistic movement primitives

    Joao Carvalho, Dorothea Koert, Marek Daniv, and Jan Peters. Residual robot learning for object-centric probabilistic movement primitives. arXiv preprint arXiv:2203.03918, 2022

    arXiv 2022

Show all 113 references
  1. [9]

    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 897...

    2019

  2. [10]

    Expressive whole- body control for humanoid robots

    Xuxin Cheng, Yandong Ji, Junming Chen, Ruihan Yang, Ge Yang, and Xiaolong Wang. Expressive whole- body control for humanoid robots. arXiv preprint arXiv:2402.16796, 2024

    2024 arXiv

  3. [11]

    Iterative residual policy: for goal-conditioned dynamic manipulation of deformable objects

    Cheng Chi, Benjamin Burchfiel, Eric Cousineau, Siyuan Feng, and Shuran Song. Iterative residual policy: for goal-conditioned dynamic manipulation of deformable objects. The International Journal of Robotics Research, 43(4):389–404, 2024

    2024

  4. [12]

    Residual learning from demonstration: Adapting dmps for contact-rich manipulation

    Todor Davchev, Kevin Sebastian Luck, Michael Burke, Franziska Meier, Stefan Schaal, and Subramanian Ra- mamoorthy. Residual learning from demonstration: Adapting dmps for contact-rich manipulation. IEEE Robotics and Automation Letters, 7(2):4488–4495, 2022

    2022

  5. [13]

    Auto-tuned sim-to-real transfer

    Yuqing Du, Olivia Watkins, Trevor Darrell, Pieter Abbeel, and Deepak Pathak. Auto-tuned sim-to-real transfer. In 2021 IEEE International Conference on Robotics and Automation (ICRA), pages 1290–1296. IEEE, 2021

    2021

  6. [14]

    Deep whole-body control: learning a unified policy for ma- nipulation and locomotion

    Zipeng Fu, Xuxin Cheng, and Deepak Pathak. Deep whole-body control: learning a unified policy for ma- nipulation and locomotion. In Conference on Robot Learning, pages 138–149. PMLR, 2023

    2023

  7. [15]

    Humanplus: Humanoid shad- owing and imitation from humans

    Zipeng Fu, Qingqing Zhao, Qi Wu, Gordon Wet- zstein, and Chelsea Finn. Humanplus: Humanoid shad- owing and imitation from humans. arXiv preprint arXiv:2406.10454, 2024

    2024 arXiv

  8. [16]

    Neu- ral internal model control: Learning a robust control policy via predictive error feedback

    Feng Gao, Chao Yu, Yu Wang, and Yi Wu. Neu- ral internal model control: Learning a robust control policy via predictive error feedback. arXiv preprint arXiv:2411.13079, 2024

    2024 arXiv

  9. [17]

    Coohoi: Learning cooperative human- object interaction with manipulated object dynamics

    Jiawei Gao, Ziqin Wang, Zeqi Xiao, Jingbo Wang, Tai Wang, Jinkun Cao, Xiaolin Hu, Si Liu, Jifeng Dai, and Jiangmiao Pang. Coohoi: Learning cooperative human- object interaction with manipulated object dynamics. arXiv preprint arXiv:2406.14558, 2024

    2024 arXiv

  10. [18]

    Dynamic identification of a 6 dof robot without joint position data

    Maxime Gautier, Pierre-Olivier Vandanjon, and Alexan- dre Janot. Dynamic identification of a 6 dof robot without joint position data. In 2011 IEEE International Conference on Robotics and Automation, pages 234–

    2011

  11. [19]

    Ad- vancing humanoid locomotion: Mastering challenging terrains with denoising world model learning

    Xinyang Gu, Yen-Jen Wang, Xiang Zhu, Chengming Shi, Yanjiang Guo, Yichen Liu, and Jianyu Chen. Ad- vancing humanoid locomotion: Mastering challenging terrains with denoising world model learning. arXiv preprint arXiv:2408.14472, 2024

    2024 arXiv

  12. [20]

    Teach a robot to fish: Versatile imitation from one minute of demonstrations

    Siddhant Haldar, Jyothish Pari, Anant Rai, and Ler- rel Pinto. Teach a robot to fish: Versatile imitation from one minute of demonstrations. arXiv preprint arXiv:2303.01497, 2023

    2023 arXiv

  13. [21]

    An iterative approach for accurate dynamic model iden- tification of industrial robots

    Yong Han, Jianhua Wu, Chao Liu, and Zhenhua Xiong. An iterative approach for accurate dynamic model iden- tification of industrial robots. IEEE Transactions on Robotics, 36(5):1577–1594, 2020

    2020

  14. [22]

    Syn- thesizing physical character-scene interactions

    Mohamed Hassan, Yunrong Guo, Tingwu Wang, Michael Black, Sanja Fidler, and Xue Bin Peng. Syn- thesizing physical character-scene interactions. In ACM SIGGRAPH 2023 Conference Proceedings, pages 1–9, 2023

    2023

  15. [23]

    Self- supervised meta-learning for all-layer dnn-based adap- tive control with stability guarantees

    Guanqi He, Yogita Choudhary, and Guanya Shi. Self- supervised meta-learning for all-layer dnn-based adap- tive control with stability guarantees. arXiv preprint arXiv:2410.07575, 2024

    2024 arXiv

  16. [24]

    Omnih2o: Universal and dexterous human- to-humanoid whole-body teleoperation and learning

    Tairan He, Zhengyi Luo, Xialin He, Wenli Xiao, Chong Zhang, Weinan Zhang, Kris Kitani, Changliu Liu, and Guanya Shi. Omnih2o: Universal and dexterous human- to-humanoid whole-body teleoperation and learning. arXiv preprint arXiv:2406.08858, 2024

    2024 arXiv

  17. [25]

    Learn- ing human-to-humanoid real-time whole-body teleop- eration

    Tairan He, Zhengyi Luo, Wenli Xiao, Chong Zhang, Kris Kitani, Changliu Liu, and Guanya Shi. Learn- ing human-to-humanoid real-time whole-body teleop- eration. arXiv preprint arXiv:2403.04436, 2024

    2024 arXiv

  18. [26]

    Hover: Versatile neural whole-body controller for humanoid robots

    Tairan He, Wenli Xiao, Toru Lin, Zhengyi Luo, Zhenjia Xu, Zhenyu Jiang, Jan Kautz, Changliu Liu, Guanya Shi, Xiaolong Wang, et al. Hover: Versatile neural whole-body controller for humanoid robots. arXiv preprint arXiv:2410.21229, 2024

    arXiv

  19. [27]

    Probabilistic inference of simulation parameters via parallel differ- entiable simulation

    Eric Heiden, Christopher E Denniston, David Millard, Fabio Ramos, and Gaurav S Sukhatme. Probabilistic inference of simulation parameters via parallel differ- entiable simulation. In 2022 International Conference on Robotics and Automation (ICRA), pages 3638–3645. IEEE, 2022

    2022

  20. [28]

    Datt: Deep adaptive trajec- tory tracking for quadrotor control

    Kevin Huang, Rwik Rana, Alexander Spitzer, Guanya Shi, and Byron Boots. Datt: Deep adaptive trajec- tory tracking for quadrotor control. arXiv preprint arXiv:2310.09053, 2023

    2023 arXiv

  21. [29]

    Learning agile and dynamic motor skills for legged robots

    Jemin Hwangbo, Joonho Lee, Alexey Dosovitskiy, Dario Bellicoso, Vassilios Tsounis, Vladlen Koltun, and Marco Hutter. Learning agile and dynamic motor skills for legged robots. Science Robotics, 4(26):eaau5872, 2019

    2019

  22. [30]

    A generic instrumental variable approach for industrial robot identification

    Alexandre Janot, Pierre-Olivier Vandanjon, and Maxime Gautier. A generic instrumental variable approach for industrial robot identification. IEEE Transactions on Control Systems Technology, 22(1):132–145, 2013

    2013

  23. [31]

    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, 2022

    2022

  24. [32]

    Exbody2: Advanced expressive humanoid whole-body control

    Mazeyu Ji, Xuanbin Peng, Fangchen Liu, Jialong Li, Ge Yang, Xuxin Cheng, and Xiaolong Wang. Exbody2: Advanced expressive humanoid whole-body control. arXiv preprint arXiv:2412.13196, 2024

    2024 arXiv

  25. [33]

    Transic: Sim-to-real policy transfer by learning from online correction

    Yunfan Jiang, Chen Wang, Ruohan Zhang, Jiajun Wu, and Li Fei-Fei. Transic: Sim-to-real policy transfer by learning from online correction. arXiv preprint arXiv:2405.10315, 2024

    2024 arXiv

  26. [34]

    Residual reinforcement learning for robot control

    Tobias Johannink, Shikhar Bahl, Ashvin Nair, Jianlan Luo, Avinash Kumar, Matthias Loskyll, Juan Aparicio Ojea, Eugen Solowjow, and Sergey Levine. Residual reinforcement learning for robot control. In 2019 international conference on robotics and automation (ICRA), pages 6023–6...

    2019

  27. [35]

    Reinforced grounded action transformation for sim-to-real transfer

    Haresh Karnan, Siddharth Desai, Josiah P Hanna, Gar- rett Warnell, and Peter Stone. Reinforced grounded action transformation for sim-to-real transfer. In 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 4397–4402. IEEE, 2020

    2020

  28. [36]

    Parameter identification of robot dynamics

    Pradeep K Khosla and Takeo Kanade. Parameter identification of robot dynamics. In 1985 24th IEEE conference on decision and control, pages 1754–1760. IEEE, 1985

    1985

  29. [37]

    Dy- namic locomotion for passive-ankle biped robots and humanoids using whole-body locomotion control

    Donghyun Kim, Steven Jens Jorgensen, Jaemin Lee, Junhyeok Ahn, Jianwen Luo, and Luis Sentis. Dy- namic locomotion for passive-ankle biped robots and humanoids using whole-body locomotion control. Int. J. Robotics Res., 39(8), 2020. doi: 10.1177/027836492091

    2020 doi

  30. [38]

    Com- bining learned and analytical models for predicting action effects from sensory data

    Alina Kloss, Stefan Schaal, and Jeannette Bohg. Com- bining learned and analytical models for predicting action effects from sensory data. The International Journal of Robotics Research, 41(8):778–797, 2022

    2022

  31. [39]

    System identification tech- niques

    F Kozin and HG Natke. System identification tech- niques. Structural safety, 3(3-4):269–316, 1986

    1986

  32. [40]

    Rma: Rapid motor adaptation for legged robots

    Ashish Kumar, Zipeng Fu, Deepak Pathak, and Jitendra Malik. Rma: Rapid motor adaptation for legged robots. arXiv preprint arXiv:2107.04034, 2021

    2021 arXiv

  33. [41]

    Adapting rapid motor adaptation for bipedal robots

    Ashish Kumar, Zhongyu Li, Jun Zeng, Deepak Pathak, Koushil Sreenath, and Jitendra Malik. Adapting rapid motor adaptation for bipedal robots. In 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 1161–1168. IEEE, 2022

    2022

  34. [42]

    Learning quadrupedal locomotion over challenging terrain

    Joonho Lee, Jemin Hwangbo, Lorenz Wellhausen, Vladlen Koltun, and Marco Hutter. Learning quadrupedal locomotion over challenging terrain. Science robotics, 5(47):eabc5986, 2020

    2020

  35. [43]

    Pi-ars: Acceler- ating evolution-learned visual-locomotion with predic- tive information representations

    Kuang-Huei Lee, Ofir Nachum, Tingnan Zhang, Sergio Guadarrama, Jie Tan, and Wenhao Yu. Pi-ars: Acceler- ating evolution-learned visual-locomotion with predic- tive information representations. In 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), ...

    2022

  36. [44]

    Using deep reinforcement learning to learn high-level policies on the atrias biped

    Tianyu Li, Hartmut Geyer, Christopher G Atkeson, and Akshara Rai. Using deep reinforcement learning to learn high-level policies on the atrias biped. In 2019 International Conference on Robotics and Automation (ICRA), pages 263–269. IEEE, 2019

    2019

  37. [45]

    Reinforcement learning for robust parameter- ized locomotion control of bipedal robots

    Zhongyu Li, Xuxin Cheng, Xue Bin Peng, Pieter Abbeel, Sergey Levine, Glen Berseth, and Koushil Sreenath. Reinforcement learning for robust parameter- ized locomotion control of bipedal robots. In 2021 IEEE International Conference on Robotics and Automation (ICRA), pages 2811–...

    2021

  38. [46]

    Robust and versatile bipedal jumping control through reinforcement learning

    Zhongyu Li, Xue Bin Peng, Pieter Abbeel, Sergey Levine, Glen Berseth, and Koushil Sreenath. Robust and versatile bipedal jumping control through reinforcement learning. arXiv preprint arXiv:2302.09450, 2023

    2023 arXiv

  39. [47]

    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

  40. [48]

    Berkeley humanoid: A research platform for learning-based con- trol

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

    2024 arXiv

  41. [49]

    System identification

    Lennart Ljung. System identification. In Signal analysis and prediction, pages 163–173. Springer

  42. [50]

    Learning humanoid locomotion with perceptive internal model

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

    2024 arXiv

  43. [51]

    Hybrid internal model: Learning agile legged locomotion with simulated robot response

    Junfeng Long, Zirui Wang, Quanyi Li, Liu Cao, Jiawei Gao, and Jiangmiao Pang. Hybrid internal model: Learning agile legged locomotion with simulated robot response. In The Twelfth International Conference on Learning Representations, 2024

    2024

  44. [52]

    Smpl: A skinned multi-person linear model

    Matthew Loper, Naureen Mahmood, Javier Romero, Gerard Pons-Moll, and Michael J Black. Smpl: A skinned multi-person linear model. In Seminal Graphics Papers: Pushing the Boundaries, V olume2, pages 851–

  45. [53]

    Mobile-television: Predictive motion priors for humanoid whole-body control

    Chenhao Lu, Xuxin Cheng, Jialong Li, Shiqi Yang, Mazeyu Ji, Chengjing Yuan, Ge Yang, Sha Yi, and Xiaolong Wang. Mobile-television: Predictive motion priors for humanoid whole-body control. arXiv preprint arXiv:2412.07773, 2024

    2024 arXiv

  46. [54]

    Pie: Parkour with implicit-explicit learning framework for legged robots

    Shixin Luo, Songbo Li, Ruiqi Yu, Zhicheng Wang, Jun Wu, and Qiuguo Zhu. Pie: Parkour with implicit-explicit learning framework for legged robots. IEEE Robotics and Automation Letters, 2024

    2024

  47. [55]

    Perpetual humanoid control for real-time sim- ulated avatars

    Zhengyi Luo, Jinkun Cao, Kris Kitani, Weipeng Xu, et al. Perpetual humanoid control for real-time sim- ulated avatars. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 10895–10904, 2023

    2023

  48. [56]

    Smplolympics: Sports environments for physically simulated hu- manoids

    Zhengyi Luo, Jiashun Wang, Kangni Liu, Haotian Zhang, Chen Tessler, Jingbo Wang, Ye Yuan, Jinkun Cao, Zihui Lin, Fengyi Wang, et al. Smplolympics: Sports environments for physically simulated hu- manoids. arXiv preprint arXiv:2407.00187, 2024

    2024 arXiv

  49. [57]

    Rl2ac: Reinforce- ment learning-based rapid online adaptive control for legged robot robust locomotion

    Shangke Lyu, Xin Lang, Han Zhao, Hongyin Zhang, Pengxiang Ding, and Donglin Wang. Rl2ac: Reinforce- ment learning-based rapid online adaptive control for legged robot robust locomotion. In Proceedings of the Robotics: Science and Systems, 2024

    2024

  50. [58]

    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. In Thirty-fifth Conference on Neural Info...

    2021

  51. [59]

    Rapid locomotion via rein- forcement learning

    Gabriel B Margolis, Ge Yang, Kartik Paigwar, Tao Chen, and Pulkit Agrawal. Rapid locomotion via rein- forcement learning. arXiv preprint arXiv:2205.02824, 2022

    2022 arXiv

  52. [60]

    Learning to see physical properties with active sensing motor policies

    Gabriel B Margolis, Xiang Fu, Yandong Ji, and Pulkit Agrawal. Learning to see physical properties with active sensing motor policies. arXiv preprint arXiv:2311.01405, 2023

    2023 arXiv

  53. [61]

    Rapid locomotion via reinforcement learning

    Gabriel B Margolis, Ge Yang, Kartik Paigwar, Tao Chen, and Pulkit Agrawal. Rapid locomotion via reinforcement learning. The International Journal of Robotics Research, 43(4):572–587, 2024

    2024

  54. [62]

    Learn- ing robust perceptive locomotion for quadrupedal robots in the wild

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

    2022

  55. [63]

    Orbit: A unified simulation framework for interactive robot learning environments

    Mayank Mittal, Calvin Yu, Qinxi Yu, Jingzhou Liu, Nikita Rudin, David Hoeller, Jia Lin Yuan, Ritvik Singh, Yunrong Guo, Hammad Mazhar, Ajay Man- dlekar, Buck Babich, Gavriel State, Marco Hutter, and Animesh Garg. Orbit: A unified simulation framework for interactive robot lear...

    2023

  56. [64]

    Data-efficient domain randomization with bayesian optimization

    Fabio Muratore, Christian Eilers, Michael Gienger, and Jan Peters. Data-efficient domain randomization with bayesian optimization. IEEE Robotics and Automation Letters, 6(2):911–918, 2021

    2021

  57. [65]

    Dreamwaq: Learning robust quadrupedal lo- comotion with implicit terrain imagination via deep reinforcement learning

    I Made Aswin Nahrendra, Byeongho Yu, and Hyun Myung. Dreamwaq: Learning robust quadrupedal lo- comotion with implicit terrain imagination via deep reinforcement learning. In 2023 IEEE International Conference on Robotics and Automation (ICRA), pages 5078–5084. IEEE, 2023

    2023

  58. [66]

    Neural-fly enables rapid learning for agile flight in strong winds

    Michael O’Connell, Guanya Shi, Xichen Shi, Kamyar Azizzadenesheli, Anima Anandkumar, Yisong Yue, and Soon-Jo Chung. Neural-fly enables rapid learning for agile flight in strong winds. Science Robotics, 7(66): eabm6597, 2022

    2022

  59. [67]

    Deepmimic: Example-guided deep reinforcement learning of physics-based character skills

    Xue Bin Peng, Pieter Abbeel, Sergey Levine, and Michiel Van de Panne. Deepmimic: Example-guided deep reinforcement learning of physics-based character skills. ACM Transactions On Graphics (TOG), 37(4): 1–14, 2018

    2018

  60. [68]

    Sim-to-real transfer of robotic control with dynamics randomization

    Xue Bin Peng, Marcin Andrychowicz, Wojciech Zaremba, and Pieter Abbeel. Sim-to-real transfer of robotic control with dynamics randomization. In 2018 IEEE international conference on robotics and automation (ICRA), pages 3803–3810. IEEE, 2018

    2018

  61. [69]

    Sfv: Reinforcement learn- ing of physical skills from videos

    Xue Bin Peng, Angjoo Kanazawa, Jitendra Malik, Pieter Abbeel, and Sergey Levine. Sfv: Reinforcement learn- ing of physical skills from videos. ACM Transactions On Graphics (TOG), 37(6):1–14, 2018

    2018

  62. [70]

    Learning agile robotic locomotion skills by imitating animals

    Xue Bin Peng, Erwin Coumans, Tingnan Zhang, Tsang- Wei Lee, Jie Tan, and Sergey Levine. Learning agile robotic locomotion skills by imitating animals. arXiv preprint arXiv:2004.00784, 2020

    2004 arXiv

  63. [71]

    Ase: Large-scale reusable adversarial skill embeddings for physically simulated characters

    Xue Bin Peng, Yunrong Guo, Lina Halper, Sergey Levine, and Sanja Fidler. Ase: Large-scale reusable adversarial skill embeddings for physically simulated characters. ACM Transactions On Graphics (TOG), 41 (4):1–17, 2022

    2022

  64. [72]

    In-hand object rotation via rapid motor adaptation

    Haozhi Qi, Ashish Kumar, Roberto Calandra, Yi Ma, and Jitendra Malik. In-hand object rotation via rapid motor adaptation. In Conference on Robot Learning, pages 1722–1732. PMLR, 2023

    2023

  65. [73]

    Humanoid locomotion as next token prediction

    I Radosavovic, B Zhang, B Shi, J Rajasegaran, S Ka- mat, T Darrell, K Sreenath, and J Malik. Humanoid locomotion as next token prediction. arxiv. 2024. arXiv preprint arXiv:2402.19469, 2024

    2024 arXiv

  66. [74]

    Real- world humanoid locomotion with reinforcement learn- ing

    Ilija Radosavovic, Tete Xiao, Bike Zhang, Trevor Dar- rell, Jitendra Malik, and Koushil Sreenath. Real- world humanoid locomotion with reinforcement learn- ing. Science Robotics, 9(89):eadi9579, 2024

    2024

  67. [75]

    Bayessim: adaptive domain randomization via probabilistic inference for robotics simulators

    Fabio Ramos, Rafael Carvalhaes Possas, and Dieter Fox. Bayessim: adaptive domain randomization via probabilistic inference for robotics simulators. arXiv preprint arXiv:1906.01728, 2019

    1906 arXiv

  68. [76]

    Adaptsim: Task-driven simula- tion adaptation for sim-to-real transfer

    Allen Z Ren, Hongkai Dai, Benjamin Burchfiel, and Anirudha Majumdar. Adaptsim: Task-driven simula- tion adaptation for sim-to-real transfer. arXiv preprint arXiv:2302.04903, 2023

    2023 arXiv

  69. [77]

    Unitree g1 humanoid agent ai avatar,

    Unitree Robotics. Unitree g1 humanoid agent ai avatar,

  70. [78]

    Unitree h1 the world’s first full-size motor drive humanoid robot flips on ground, 2024

    Unitree Robotics. Unitree h1 the world’s first full-size motor drive humanoid robot flips on ground, 2024. URL https://www.youtube.com/watch?v=V1LyWsiTgms

    2024

  71. [79]

    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

  72. [80]

    URL https://www.unitree.com/g1

  73. [81]

    Neural lander: Stable drone landing control using learned dynam- ics

    Guanya Shi, Xichen Shi, Michael O’Connell, Rose Yu, Kamyar Azizzadenesheli, Animashree Anandku- mar, Yisong Yue, and Soon-Jo Chung. Neural lander: Stable drone landing control using learned dynam- ics. In 2019 international conference on robotics and automation (icra), pages 9...

    2019

  74. [82]

    Neural-swarm2: Planning and con- trol of heterogeneous multirotor swarms using learned interactions

    Guanya Shi, Wolfgang H ¨onig, Xichen Shi, Yisong Yue, and Soon-Jo Chung. Neural-swarm2: Planning and con- trol of heterogeneous multirotor swarms using learned interactions. IEEE Transactions on Robotics, 38(2): 1063–1079, 2021

    2021

  75. [83]

    Proximal policy optimiza- tion algorithms

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

    2017 arXiv

  76. [84]

    Residual policy learning

    Tom Silver, Kelsey Allen, Josh Tenenbaum, and Leslie Kaelbling. Residual policy learning. arXiv preprint arXiv:1812.06298, 2018

    2018 arXiv

  77. [85]

    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

  78. [86]

    Pip-loco: A proprioceptive infinite horizon planning framework for quadrupedal robot locomotion

    Aditya Shirwatkar, Naman Saxena, Kishore Chandra, and Shishir Kolathaya. Pip-loco: A proprioceptive infinite horizon planning framework for quadrupedal robot locomotion. arXiv preprint arXiv:2409.09441, 2024

    2024 arXiv

  79. [87]

    Domain random- ization for transferring deep neural networks from simu- lation to the real world

    Josh Tobin, Rachel Fong, Alex Ray, Jonas Schneider, Wojciech Zaremba, and Pieter Abbeel. Domain random- ization for transferring deep neural networks from simu- lation to the real world. In 2017 IEEE/RSJ international conference on intelligent robots and systems (IROS), pages ...

    2017

  80. [88]

    Mu- joco: A physics engine for model-based control

    Emanuel Todorov, Tom Erez, and Yuval Tassa. Mu- joco: A physics engine for model-based control. In 2012 IEEE/RSJ International Conference on Intelligent Robots and Systems, pages 5026–5033. IEEE, 2012. doi: 10.1109/IROS.2012.6386109

    2012

  81. [89]

    Maskedmimic: Unified physics- based character control through masked motion inpaint- ing

    Chen Tessler, Yunrong Guo, Ofir Nabati, Gal Chechik, and Xue Bin Peng. Maskedmimic: Unified physics- based character control through masked motion inpaint- ing. ACM Transactions on Graphics (TOG), 43(6):1–21, 2024

    2024

  82. [90]

    Strategy and skill learning for physics-based table tennis animation

    Jiashun Wang, Jessica Hodgins, and Jungdam Won. Strategy and skill learning for physics-based table tennis animation. In ACM SIGGRAPH 2024 Conference Papers, pages 1–11, 2024

    2024

  83. [91]

    Physhoi: Physics-based imitation of dynamic human-object interaction

    Yinhuai Wang, Jing Lin, Ailing Zeng, Zhengyi Luo, Jian Zhang, and Lei Zhang. Physhoi: Physics-based imitation of dynamic human-object interaction. arXiv preprint arXiv:2312.04393, 2023

    2023 arXiv

  84. [92]

    Cts: Concurrent teacher-student reinforcement learning for legged locomotion

    Hongxi Wang, Haoxiang Luo, Wei Zhang, and Hua Chen. Cts: Concurrent teacher-student reinforcement learning for legged locomotion. IEEE Robotics and Automation Letters, 2024

    2024

  85. [93]

    Tram: Global trajectory and motion of 3d humans from in-the-wild videos

    Yufu Wang, Ziyun Wang, Lingjie Liu, and Kostas Dani- ilidis. Tram: Global trajectory and motion of 3d humans from in-the-wild videos. In European Conference on Computer Vision, pages 467–487. Springer, 2025

    2025

  86. [94]

    Toward understanding key estimation in learning robust humanoid locomotion

    Zhicheng Wang, Wandi Wei, Ruiqi Yu, Jun Wu, and Qiuguo Zhu. Toward understanding key estimation in learning robust humanoid locomotion. arXiv preprint arXiv:2403.05868, 2024

    2024 arXiv

  87. [95]

    Skillmimic: Learning reusable basketball skills from demonstrations

    Yinhuai Wang, Qihan Zhao, Runyi Yu, Ailing Zeng, Jing Lin, Zhengyi Luo, Hok Wai Tsui, Jiwen Yu, Xiu Li, Qifeng Chen, et al. Skillmimic: Learning reusable basketball skills from demonstrations. arXiv preprint arXiv:2408.15270, 2024

    2024 arXiv

  88. [96]

    Loopsr: Looping sim-and-real for life- long policy adaptation of legged robots

    Peilin Wu, Weiji Xie, Jiahang Cao, Hang Lai, and Weinan Zhang. Loopsr: Looping sim-and-real for life- long policy adaptation of legged robots. arXiv preprint arXiv:2409.17992, 2024

    2024

  89. [97]

    Anycar to anywhere: Learning universal dynamics model for agile and adaptive mobil- ity

    Wenli Xiao, Haoru Xue, Tony Tao, Dvij Kalaria, John M Dolan, and Guanya Shi. Anycar to anywhere: Learning universal dynamics model for agile and adaptive mobil- ity. arXiv preprint arXiv:2409.15783, 2024

    2024 arXiv

  90. [98]

    Ai for humanoid robotics - a lecture by mentee robotics’ ceo, prof

    Lior Wolf. Ai for humanoid robotics - a lecture by mentee robotics’ ceo, prof. lior wolf. YouTube video,

  91. [99]

    Agile continuous jumping in discontinuous terrains

    Yuxiang Yang, Guanya Shi, Changyi Lin, Xiangyun Meng, Rosario Scalise, Mateo Guaman Castro, Wenhao Yu, Tingnan Zhang, Ding Zhao, Jie Tan, et al. Agile continuous jumping in discontinuous terrains. arXiv preprint arXiv:2409.10923, 2024

    2024 arXiv

  92. [100]

    Preparing for the unknown: Learning a universal pol- icy with online system identification

    Wenhao Yu, Jie Tan, C Karen Liu, and Greg Turk. Preparing for the unknown: Learning a universal pol- icy with online system identification. arXiv preprint arXiv:1702.02453, 2017

    2017 arXiv

  93. [101]

    Policy transfer with strategy optimization

    Wenhao Yu, C Karen Liu, and Greg Turk. Policy transfer with strategy optimization. arXiv preprint arXiv:1810.05751, 2018

    2018 arXiv

  94. [102]

    Learning locomotion skills for cassie: Iterative design and sim-to- real

    Zhaoming Xie, Patrick Clary, Jeremy Dao, Pedro Morais, Jonanthan Hurst, and Michiel Panne. Learning locomotion skills for cassie: Iterative design and sim-to- real. In Conference on Robot Learning, pages 317–329. PMLR, 2020

    2020

  95. [103]

    Learning fast adaptation with meta strategy optimization

    Wenhao Yu, Jie Tan, Yunfei Bai, Erwin Coumans, and Sehoon Ha. Learning fast adaptation with meta strategy optimization. IEEE Robotics and Automation Letters, 5(2):2950–2957, 2020

    2020

  96. [104]

    Learning phys- ically simulated tennis skills from broadcast videos

    YE YUAN and Viktor Makoviychuk. Learning phys- ically simulated tennis skills from broadcast videos. 2023

    2023

  97. [105]

    Wococo: Learning whole-body humanoid control with sequential contacts

    Chong Zhang, Wenli Xiao, Tairan He, and Guanya Shi. Wococo: Learning whole-body humanoid control with sequential contacts. arXiv preprint arXiv:2406.06005, 2024

    2024 arXiv

  98. [106]

    Sim-to-real transfer for biped locomotion

    Wenhao Yu, Visak CV Kumar, Greg Turk, and C Karen Liu. Sim-to-real transfer for biped locomotion. In 2019 ieee/rsj international conference on intelligent robots and systems (iros), pages 3503–3510. IEEE, 2019

    2019

  99. [107]

    Humanoid parkour learning

    Ziwen Zhuang, Shenzhe Yao, and Hang Zhao. Humanoid parkour learning. arXiv preprint arXiv:2406.10759, 2024. APPENDIX A. Domain Randomization in Pre-Training To improve the robustness and generalization of the pre- trained policy in Figure 2 (a), we utilized the domain random- ...

    2024 arXiv

  100. [110]

    Whole-body humanoid robot locomotion with human reference

    Qiang Zhang, Peter Cui, David Yan, Jingkai Sun, Yiqun Duan, Gang Han, Wen Zhao, Weining Zhang, Yijie Guo, Arthur Zhang, et al. Whole-body humanoid robot locomotion with human reference. In 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 1...

    2024

  101. [112]

    Fixed-Point Iteration : We initialize y0 = ˆπ(s) and iteratively update: yk+1 = ˆπ(s) − π∆(s, yk), (3) where yk converges to a solution after K iterations

  102. [113]

    (4) A gradient descent method minimizes this loss to solve for y

    Gradient-Based Optimization : Define the loss func- tion: l(y) = ∥y + π∆(s, y) − ˆπ(s)∥2. (4) A gradient descent method minimizes this loss to solve for y. These methods approximate π(s), but suffer from OOD issues when trained on limited trajectories. RL fine-tuning, in contr...

  103. [2024]

    URL https://github.com/Genesis-Embodied-AI/ Genesis

  104. [2025]

    Accessed: 2025-01-31

    URL https://www.youtube.com/watch?v=y1LG 4YwUtoo. Accessed: 2025-01-31

    2025

  105. [8014]

    URL https://doi.org/10.1177/0278364920918014

    doi

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.