MuscleControl-Isaac

An IsaacLab-based repository for muscle-activation control and muscle/skeleton visualization. This project extends the IsaacLab ecosystem with muscle-driven human models and training/evaluation pipelines for research in biomechanical control, imitation learning, and motion generation.

Highlights

• Muscle-activation control: muscle-driven human models and control interfaces.
• Visualization: muscle and skeleton mesh rendering for analysis and debugging.
• Training & evaluation: integrated scripts and entry points for experiments.
• DiffusionForcing Control: use diffusion forcing to guide and control.

Install

1. Install IsaacLab
2. Once IsaacLab is installed, from the repository root install the ProtoMotions package and its dependencies with:

Set PYTHON_PATH to point at the isaaclab.sh script For Linux:

alias PYTHON_PATH="<isaac_lab_path> -p"

For example:

alias PYTHON_PATH="/home/USERNAME/IsaacLab/isaaclab.sh -p"

Once IsaacLab is installed, from the protomotions repository root, install the Physical Animation package and its dependencies with:

PYTHON_PATH -m pip install -e .
PYTHON_PATH -m pip install -r requirements_isaaclab.txt
PYTHON_PATH -m pip install -e isaac_utils
PYTHON_PATH -m pip install -e poselib

Training

PD(phase 1)

CUDA_VISIBLE_DEVICES=0 python protomotions/train_agent.py \
    +exp=full_body_tracker/transformer_flat_terrain \
    +robot=bio_act \
    +simulator=isaaclab \
    motion_file=./data/amass_retarget.pt \
    +experiment_name=full_body_tracker_motionfix_v2 \
    num_envs=1024 \
    agent.config.batch_size=4096 \
    agent.config.num_mini_epochs=2 \
    agent.config.eval_metrics_every=2000 \
    +opt=wandb \
    wandb.wandb_id=${WANDB_ID:-null} \
    wandb.wandb_resume=allow \
    +agent.config.train_teacher=true \
    ngpu=1

Muscle(phase 2)

CUDA_VISIBLE_DEVICES=0 python protomotions/train_agent.py \
    +exp=mus/no_vae_no_text_flat_terrain \
    +robot=bio_act_stu \
    +simulator=isaaclab \
    motion_file=./data/amass_filtered.pt \
    +experiment_name=student_2 \
    num_envs=1024 \
    agent.config.batch_size=2048 \
    agent.config.num_mini_epochs=2 \
    agent.config.eval_metrics_every=2000 \
    +opt=wandb \
    wandb.wandb_id=${WANDB_ID:-null} \
    wandb.wandb_resume=allow \
    agent.config.expert_model_path=results/full_body_tracker_motionfix_v2 \
    ngpu=1

Demos

MotionTracking

GenerativeMotion(DiffusionForcing)

Acknowledgements

Built on IsaacLab and ProtoMotions. Thanks to the community and upstream dependencies.