Model-based Contact-rich Manipulation Planning

This repo provides implementation of model-based planners for contact-rich manipulation. The attached Deepnote project includes two notebooks, illustrating two planning algorithms on the Allegro hand dexterous manipulation example:

• A trajectory optimizer using iterative MPC (iMPC),
• A sampling-based planner capable of handling contact dynamics constraints.

Details of the planning algorithms can be found in

• Global Planning for Contact-Rich Manipulation via Local Smoothing of Quasi-dynamic Contact Models, currently under review.

Our quasidynamic simulator can be found on the tro2023 branch of the quasistatic_simulator repo:

• https://github.com/pangtao22/quasistatic_simulator/tree/tro2023

To make it work locally

• download the deepnote folder
• 'cd planning-through-contact-deepnote/planning-through-contact'
• pull the branch for planning-through-contact
  • fork url https://github.com/slecleach/planning_through_contact.git
  • branch benchmark

build docker image

• cd planning-through-contact-deepnote
• docker build -t crm_image .

run docker

• cd planning-through-contact-deepnote
• docker run -it -v ~/workspace/planning-through-contact-deepnnote:/planning_through_contact crm_image
• directly in docker pip install plotly, pandas

run notebook in vscode

• attach to running container
• ctrl+shift+p, "python select interpreter", usr/bin/python
• run planar_hand_rrt.ipynb