This is the pretrained model, training data set, evaluation problem sets, and sample real robot data accompanying Motion Policy Networks, an end-to-end neural model that can be used to generate collision-free, smooth motion from just a single depth camera observation. The training data set consists of over 3 million motion planning problems for a Franka Panda arm in over 500,000 environments.

Version 1.0.3 uses the new mpinets_types.py API instead of types.py API to prevent import errors--otherwise the data is the same