An easy task (left) and a harder task (right). The robot is initialised with a position sampled from a uniform distribution centered at the demonstrated trajectory's initial position with variance of +/- 11.8cm along all dimensions. The difficulty is defined by the size of the hole. A task is complete when the peg is fully inserted.
Comparison between using residual, hybrid, DMP and model-free policies. The residual policy (green) consistently results in experiencing comparable to only using DMP (red) and even lower force across both the easy and hard tasks. Lower is better.