Paper on Subgoal Discovery accepted in NeurIPS 2019
Our recent paper, Learning from Trajectories via Subgoal Discovery, in NeurIPS 2019 proposes an approach to discover subgoals to solve complex tasks, from a limited number of expert demonstrations. Using these discovered subgoals leads to significantly faster robotic task learning with none to limited hand-designing of rewards. This work is done in collaboration between UCR and MERL (Mitsubishi Electric Research Laboratories).
Title: Learning from Trajectories via Subgoal Discovery
S. Paul, Jeroen van Baar, A. Roy-Chowdhury, Advances in Neural Information Processing Systems (NeurIPS), 2019