PDF Paper ArXiv arXiv Youtube Video Code

ASAP

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

PDF Paper ArXiv arXiv Youtube Video Code

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

Tairan He*   Jiawei Gao*   Wenli Xiao*   Yuanhang Zhang*   Zi Wang   Jiashun Wang  
Zhengyi Luo   Guanqi He   Nikhil Sobanbabu   Chaoyi Pan   Zeji Yi   Guannan Qu  
Kris Kitani   Jessica Hodgins   Linxi "Jim" Fan   Yuke Zhu   Changliu Liu   Guanya Shi  
PDF ArXiv Video Summary Code

Agile Whole-body Skills

Cristiano Ronaldo

Kobe Bryant


LeBron James

Side Jump (1.3m)


Jump Forward (0.85m)

Jump Forward (1.5m)


Forward Kick

Right Kick


APT Dance

Leg Stretch


Squat

Squat + Lean Forward


Before/After ASAP Fine-tuning

Kick (Before ASAP)

Kick (After ASAP)


LeBron James (Before ASAP)

LeBron James (After ASAP)

Abstract

Humanoid robots hold the potential for unparalleled versatility for 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 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.

Method

    There are four steps within the ASAP framework:
    1. 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;
    2. Delta Action Model Training
      Based on the real-world rollout data, we train the delta action model by minimizing the discrepancy between simulation state s_t and real-world state s^r_t;
    3. Policy Fine-tuning
      We freeze the delta action model, incorporate it into the simulator to align the real-world physics and then fine-tune the pre-trained motion tracking policy;
    4. Real-World Deployment
      Finally, we deploy the fine-tuned policy directly in the real world without the delta action model.

BibTeX

@article{he2025asap,
          title={ASAP: Aligning Simulation and Real-World Physics for Learning Agile Humanoid Whole-Body Skills},
          author={He, Tairan and Gao, Jiawei and Xiao, Wenli and Zhang, Yuanhang and Wang, Zi and Wang, Jiashun and Luo, Zhengyi and He, Guanqi and Sobanbabu, Nikhil and Pan, Chaoyi and Yi, Zeji and Qu, Guannan and Kitani, Kris and Hodgins, Jessica and Fan, Linxi "Jim" and Zhu, Yuke and Liu, Changliu and Shi, Guanya},
          journal={arXiv preprint arXiv:2502.01143},
          year={2025}
        }