Constrained Control-as-inference
The following requirements can be installed via pip
- NumPy
- PyTorch Other dependencies:
- Isaac Gym (https://developer.nvidia.com/isaac-gym)
- isaacgym-arm-envs https://github.com/UM-ARM-Lab/isaacgym-arm-envs
- pytorch_kinematics https://github.com/UM-ARM-Lab/pytorch_kinematics/tree/kinematic_hessian
Navigate to directory and install with pip install -e .
Example scripts are in the examples folder.
double_integrator_on_sphere.py will run the planner for a 3D double integrator constraint to travel on the unit sphere.
victor_table_surface.py will run the planner on a task where the robot must move the end-effector to a goal location while maintaining contact with the table.
run_victor_wrench_sim.py will run the planner on a task where the robot must turn a wrench.
run_victor_wrench_real.py Same as above, but for running on the real robot in the lab
The planning configuration files for these examples are found in config/planning_configs in .yaml format.
quadrotor_learn_to_sample.py will train a generative model for the quadrotor example
victor_table_learn_to_sample.py will train a generative model for the victor table example
The training configuration files are found in config/training_configs in .yaml format. Using these configs you can train a diffusion model or a normalizing flow model (either by max-likelihood or flow matching)
Saved models and plots are stored in data/training, and the training data for these models is stored in data/training_data.