mjlab

Identify and simulate extended friction models for servo-actuators

MuJoco Simulation on WebAssembly with Neural networks

An open-source unsupervised RL framework for humanoid control with FB/TeCH training, robot-aware motion import, and real-world teleoperation.

Train a Unitree G1 humanoid to perform a double spin kick using mjlab

One policy. Any motion. Train universal whole-body tracking in mjlab—swap clips, not checkpoints.

A collection of tasks built on mjlab

AI coding skill kit for IsaacLab migration and mjlab-native development

Integrating a custom robot (ANYmal C) with mjlab's velocity task

mjlab reinforcement learning for the BDX-R robot

A growing collection of manipulation tasks built with mjlab.

Reproducible Microduck RL experiments, policies, simulation assets, and hardware add-ons

MuGS: MuJoCo Gaussian Splatting

Post-training humanoid whole-body trackers for perceptive control tasks

RL training environments for dexhand tasks (mjlab)

Perceptive behavioral foundation models for terrain-adaptive humanoid control: motion tracking, goal reaching, reward inference and teleoperation.

Lightweight, GPU-accelerated reinforcement learning for Isaac Lab, mjlab, and Gymnasium.

Extension of RSL-RL for using Morphological Symmetries in IsaacLab

Predictive Style Matching (PSM), is method in which an offline predictor maps the robot’s lower-body state history and velocity commands to interpretable upper-body joint and gait targets that shape the rewards during training.

Reinforcement learning for Booster T1 and K1 humanoids: velocity control, motion tracking, and AMP-based target-location locomotion. Built on MJLab.

Extension of RSL-RL for using Mixture-of-Experts in IsaacLab