A lightweight suite of motion imitation methods for training controllers.
-
Updated
Nov 18, 2026 - Python
A lightweight suite of motion imitation methods for training controllers.
[CVPR 2022] PoseTriplet: Co-evolving 3D Human Pose Estimation, Imitation, and Hallucination under Self-supervision (Oral)
[ICLR 2023] DiffMimic: Efficient Motion Mimicking with Differentiable Physics https://arxiv.org/abs/2304.03274
A Multi-Task Dataset for Simulated Humanoid Control
Biped robot locomotion with deep RL: Isaac Lab + MuJoCo, custom PPO/SAC/DDPG/D4PG/MPO, MLP/LSTM/Transformer, motion imitation, sim-to-real
A high-performance GPU framework for learning physics-based character skills in seconds, with coding-agent workflows for training, tuning and porting environments (SIGGRAPH Asia 2026).
Official implementation of "Learning to Assist: Physics-Grounded Human-Human Control via Multi-Agent Reinforcement Learning" (CVPR 2026)
From boxing to whole-body motion: compact Unitree G1 student policies in MuJoCo, with BC + DAgger + PPO, real rollout videos, checkpoints, and bilingual technical reports.
Teaching a simulated Unitree G1 to walk, cartwheel, backflip and get up with RL (mjlab on a DGX Spark) — a zero-background 15-chapter guide.
SuperMimic trained from scratch on Apple M5: MPS learning + CPU MuJoCo, all 40 LAFAN1 clips, reproducible setup, checkpoints, videos and measured results.
Extract and retarget FBX/Mixamo animations to robot joint trajectories for use as motion references in reinforcement learning.
Reference-conditioned motion imitation for Unitree AS2: simulated forward jump, experimental backflip, and single-jump sim-to-real.
RPO 23-DoF 跑步与后空翻仿真实验,基于 Motphys/Uni-SuperMimic;含演示、训练配置及框架与数据出处。
Deep Reinforcement Learning framework for Unitree G1 Humanoid (23-DOF) learning multi-motion imitation under a single unified policy in MuJoCo.
Unitree G1 humanoid learns regulation basketball shooting in Isaac Lab: PPO + human mocap motion prior (AMP) + domain randomization
Reproducing SuperMimic (arXiv:2610.28480) step by step: L0 Pendulum sanity check -> L1 MimicKit PPO baseline -> L2 three-stage training loop
HMP: Learning reusable Hybrid Motion Priors for humanoid locomotion from motion imitation (Unitree G1, mjlab).
A community-driven index of robot motion, control, imitation, locomotion, and manipulation models across brands.
LimX Dynamics HU_D03 humanoid locomotion and dance motion-tracking (mimic) policies trained with RL in mjlab (MuJoCo Lab) on a free Colab/Kaggle GPU
코드 기반 휴머노이드 모델과 AMP 동작 학습 — 걷기·춤·백덤블링 정책 연결 및 평가 (Isaac Lab 시뮬레이션)
To associate your repository with the motion-imitation topic, visit your repo's landing page and select "manage topics."