Continual Learning

We show that storing skill policies, careful pre-training, and fine-tuning with on-policy re-inforcement learning (RL), are sufficient for building a continual manipulation skill learner.