Tutorials

Notebooks live in the repository tutorials/ directory, not under docs/source/. Index and install notes: tutorials/README.md.

Notebooks are numbered <track>.<number>; extra files of X.Y live in tutorials/files/X.Y/.

  1. Basics — 1.1_Get_Started (create an env and step it), 1.2_Load_Policy (run an mjlab or SB3 checkpoint)

  2. Training — 2.1_Train_SB3_Policy (PPO, CPU), 2.2_Train_MjLab_Policy (GPU), 2.3_SAR, 2.4_MyoReflex_Walk

  3. Analysis — 3.1_Analyse_Movements, 3.2_Inverse_Kinematics, 3.3_Inverse_Dynamics, 3.4_Computed_Muscle_Control, 3.5_Playback_Mot_File

  4. Modelling and conditions — 4.1_Move_Hand_Fingers, 4.2_Fatigue_Modeling, 4.3_Modular_Task_Config

  5. MuscleMimic — 5.1_Fullbody_Load_Policy, 5.2_Fullbody_Train_Policy, 5.3_Fullbody_Train_MjLab_Policy, 5.4_MuscleMimic_Directional_Locomotion, 5.5_MuscleMimic_SAR

Muscle-condition IDs:

  • Fatigue: myoFatiElbowPose1D6MRandom-v0 (not …Fatigue…)

  • Sarcopenia: myoSarcElbowPose1D6MRandom-v0

  • Tendon transfer (hands): myoReafHandKeyTurnFixed-v0

GPU training (NVIDIA GPU, Linux or Windows):

python scripts/train_mjlab.py myoElbowPose1D6MRandom-v0 --env.scene.num-envs 1024

Add --video True to record videos during training (it renders offscreen, so it also works on a remote, headless machine). Walk-through: tutorials/2.2_Train_MjLab_Policy.ipynb.

New CPU tasks subclass MyoGymnasiumEnv and register with registry.register_env — see docs/wiki/adding-a-new-task.md and Architecture.