gbionics/amp-rsl-rl

🔁 AMP-RSL-RL: Adversarial Motion Priors for robotic RL (PPO + motion imitation)

What it solves

AMP-RSL-RL allows humanoid agents to learn complex motor skills from motion capture data. It solves the problem of manually designing reward functions for complex movements by using adversarial imitation learning to make the agent's movements look like the expert data.

How it works

The library extends the Proximal Policy Optimization (PPO) implementation of RSL-RL by incorporating Adversarial Motion Priors (AMP). It uses a discriminator network to distinguish between the agent's movements and real motion capture data, forcing the agent to mimic the expert style while achieving its goals. It also includes symmetry-aware data augmentation and mirror-loss hooks to ensure the learned policies are consistent across symmetric robot parts.

Who it’s for

It is designed for robotics researchers and developers working on humanoid motor control and reinforcement learning.

Highlights

  • Adversarial imitation learning for motor skill acquisition.
  • Built on top of the RSL-RL PPO implementation.
  • Supports symmetry-aware data augmentation and mirror-loss regularization.
  • Compatible with motion capture datasets in .npy format.

Related

  • Project
    zitongbai/legged_lab

    A reinforcement learning extension for Isaac Lab that enables humanoid robots, such as the Unitree G1, to learn complex motions via DeepMimic and Adversarial Motion Priors.

  • Project
    PrimeIntellect-ai/verifiers

    A library for creating environments to train and evaluate LLMs, specifically designed for reinforcement learning workflows.

  • Project
    alibaba/ROLL

    An efficient scaling library for reinforcement learning with LLMs that optimizes training and inference across large GPU clusters using a distributed Ray-based architecture.

  • Project
    NovaSky-AI/SkyRL

    A modular full-stack reinforcement learning library for LLMs, providing tools for training long-horizon, multi-turn agents capable of tool use in real-world environments.

  • Project
    facebookresearch/Pearl

    A production-ready Reinforcement Learning library from Meta for building modular AI agents capable of handling complex, real-world environments with sparse feedback and dynamic action spaces.