Revo-MjLab

This repo contains tasks for Revo3 base on MjLab, Standalone BrainCo task package for mjlab.

Current migration status:

• Manager-based Dexsuite tasks are migrated into a native mjlab.tasks package.

• Registered tasks:

  • BrainCo-Dexsuite-Arm-BrainCo-Lift-v0

• Direct BrainCo in-hand tasks are not yet ported in this package.

安装

conda activate mjlab
git clone https://github.com/BrainCoTech/Revo-MjLab.git
cd Revo-MjLab/srouce/
pip install -e .

任务总览

Revo3

Task Description	Task Name	Weight	Demo
Grasp (Lift)	BrainCo-Dexsuite-Arm-BrainCo-Lift-v0	-

配置文件

Revo3

1) Grasp (Lift)

• Env config: source/BrainCo_DexHand/BrainCo_DexHand/tasks/manager_based/dexsuite/config/arm_brainco/dexsuite_arm_brain_env_cfg_grasp.py (DexsuiteArmBrainCoLiftEnvCfg)
• Base config used: source/BrainCo_DexHand/BrainCo_DexHand/tasks/manager_based/dexsuite/dexsuite_env_cfg_grasp.py

训练与演示

Train command

Training does not provide an IsaacLab-style live interactive scene viewer.

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

Play command

render by local machine

play BrainCo-Dexsuite-Arm-BrainCo-Lift-v0 \
    --wandb-run-path <user>/<project_id>/<task_id>  \
    --num-envs 1 \

render by the website

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