Residual Learning from Demonstration: Adapting DMPs for Contact-Rich Manipulation

Check out our latest IEEE RA-L publication to be presented at ICRA 2022.

ResidualLearningDemo

This paper introduces a residual learning approach to adapt dynamic movement primitives to improve generalisation in contact rich manipulation tasks, specifically looking at complicated insertion tasks like gear assembly or RJ45 network cable insertion. The paper involved researchers from the University of Edinburgh, Google X, Facebook AI research, Aalto University and Monash Robotics.

For the full paper, visit: https://arxiv.org/abs/2008.07682

For the supporting website, visit https://sites.google.com/view/rlfd/home