PEBBLE: Feedback-Efficient Interactive Reinforcement Learning via Relabeling Experience and Unsupervised Pre-training.

Kimin Lee, Laura M. Smith, Pieter Abbeel

Browse the full ICML paper archive.