README.md

Revo-MjLab

BrainCo Revo3 task package for mjlab.

This repository is a standalone mjlab task package. It registers BrainCo tasks through the mjlab.tasks entry point and is intended to be used with the official mjlab train and play commands.

Included Tasks

Robot	Task	Task ID	Demo
Revo3 arm + hand	Dexsuite lift	BrainCo-Dexsuite-Arm-BrainCo-Lift-v0
Revo3 arm + hand	Dexsuite reorient	BrainCo-Dexsuite-Arm-BrainCo-Reorient-v0	-
Revo3 right hand	In-hand rotation	BrainCo-Revo3-Right-Inhand-Rotate-v0	-

Repository Layout

Revo-MjLab/
|-- image/                  # Task demo GIFs
|-- scripts/                # Small helper scripts
`-- source/
    |-- assets/             # URDF/STL assets for arm-hand Dexsuite tasks
    |-- pyproject.toml      # mjlab task package metadata
    `-- src/brainco_mjlab_tasks/
        |-- assets/         # Package-local MJCF assets
        |-- dexsuite/       # Arm-hand lift/reorient tasks
        `-- inhand/         # Revo3 right-hand in-hand rotation task

Install

Install mjlab first by following the official mjlab documentation. Then install this task package:

cd Revo-MjLab/source
pip install -e .

List Tasks

PYTHONPATH=source/src python scripts/list_envs.py --keyword BrainCo

In an installed mjlab environment, the tasks also load through the mjlab.tasks entry point.

Train

train BrainCo-Dexsuite-Arm-BrainCo-Lift-v0 \
  --video True \
  --env.scene.num-envs 4096 \
  --gpu-ids "[0]"

train BrainCo-Revo3-Right-Inhand-Rotate-v0 \
  --headless True \
  --env.scene.num-envs 4096 \
  --gpu-ids "[0]"

Play

Native MuJoCo viewer:

play BrainCo-Dexsuite-Arm-BrainCo-Lift-v0 \
  --wandb-run-path <user>/<project>/<run> \
  --num-envs 1 \
  --viewer native

Browser viewer:

play BrainCo-Dexsuite-Arm-BrainCo-Lift-v0 \
  --wandb-run-path <user>/<project>/<run> \
  --num-envs 1 \
  --viewer viser

Scope

This repository contains simulation tasks only. It does not include sim2sim or hardware deployment code.