Opportunities and Challenges from Using Animal Videos in Reinforcement Learning fro Navigation

We investigate the use of animal videos (observations) to improve Reinforcement Learning (RL) efficiency and performance in navigation tasks with sparse rewards. Motivated by theoretical considerations, we make use of weighted policy optimization for off-policy RL and describe the main challenges when learning from animal videos. We propose solutions and test our ideas on a 2D navigation task. We show how the use of animal videos improves performance over RL algorithms that do not leverage such observations. Keywords: Learning for Control; Reinforcement Learning; Deep Learning; RL with demonstrations; Imitation Learning; Bio-inspired Control.Abstract: We investigate the use of animal videos (observations) to improve Reinforcement Learning (RL) efficiency and performance in navigation tasks with sparse rewards. Motivated by theoretical considerations, we make use of weighted policy optimization for off-policy RL and describe the main challenges when learning from animal videos. We propose solutions and test our ideas on a 2D navigation task. We show how the use of animal videos improves performance over RL algorithms that do not leverage such observations.