Aim

Implementation of differentiable physics engine for distributed motion planning in multi-agent loco-manipulation tasks.

Objectives

1. Implementation of differentiable physics engine.
2. Integration of said physics’ engine with Crocoddyl optimal control library.
3. Implementation of optimal control problem in multi-agent loco-manipulation task for collaborative carrying of objects.

Description

The goal of this project is to develop a differentiable physics engine similar to [1] [2] inside the Crocoddyl optimal control library [3]. The differentiable physics engine will help to overcome constraints imposed by the non-continuous nature of the unilateral contact model. This project will enhance the capabilities of legged robots in tasks of loco-manipulation by allowing incorporation of contact planning into the optimal control problem. This can be achieved by making the contacts continuous and differentiable. It will help formulate simpler tasks for stable, shared loco-manipulation of coupled robots [4] and fully decentralize the control architecture in legged multi-agent systems.

References

[1] Taylor A. Howell, and Simon Le Cleac'h, and Jan Brüdigam, and J. Zico Kolter, and Mac Schwager, and Zachary Manchester “Dojo: A Differentiable Physics Engine for Robotics”
[2] Keenon Werling, and Dalton Omens, and Jeongseok Lee, and Ioannis Exarchos, and C. Karen Liu “Fast and Feature-Complete Differentiable Physics for Articulated Rigid Bodies with Contact”
[3] Carlos Mastalli, and Rohan Budhiraja, and Wolfgang Merkt, and Guilhem Saurel, and Bilal Hammoud, and Maximilien Naveau, and Justin Carpentier, and Ludovic Righetti, and Sethu Vijayakumar, and Nicolas Mansard “Crocoddyl: An Efficient and Versatile Framework for Multi-Contact Optimal Control”
[4] Randall T. Fawcett, and Aaron D. Ames, and Kaveh Akbari Hamed “Distributed Planning of  Collaborative Locomotion: A Physics-Based and Data-Driven Approach”