google-deepmind / mujoco_mpc

Real-time behaviour synthesis with MuJoCo, using Predictive Control
https://github.com/deepmind/mujoco_mpc
Apache License 2.0
1.06k stars 157 forks source link

Added support for motion strategies #341

Open nkhosh opened 2 months ago

nkhosh commented 2 months ago

In the Humanoid Interact task, this PR adds:

  1. Motion strategy: a class which manages a sequence of contact keyframes, and transitions between them.
  2. Success criteria (time limit, distance threshold and success sustain time) for each contact keyframe.
  3. Ability to serialize/deserialize (save and load) motion strategies.
  4. Sample motion strategies to be readily loaded into the task.