On the Opportunities and Challenges of using Animals Videos in Reinforcement Learning
We investigate the use of animals videos to improve efficiency and performance in Reinforcement Learning (RL). Under a theoretical perspective, we motivate the use of weighted policy optimization for off-policy RL, describe the main challenges when learning from videos and propose solutions. We test our ideas in offline and online RL and show encouraging results on a series of 2D navigation tasks.
READ FULL TEXT