Task-Error Residual Learning for Real-Robot Five-Ball Juggling

Kai Ploeger, Jan Peters

For residual learning that refines existing behavior, sample efficiency depends on two things: how much information each rollout returns, and how efficiently the learner uses that…