Goal-conditioned Batch Reinforcement Learning for Rotation Invariant Locomotion

Abstract: We propose a novel approach to learn goal-conditioned policies for locomotion in a batch RL setting. The batch data is collected by a policy that is not goal-conditioned. For the locomotion task, this translates to data collection using a policy learnt by the agent for walking straight in one direction, and using that data to learn a goal-conditioned policy that enables the agent to walk in any direction. The data collection policy used should be invariant to the direction the agent is facing i.e. regardless of its initial orientation, the agent should take the same actions to walk forward. We exploit this property to learn a goal-conditioned policy using two key ideas: (1) augmenting data by generating trajectories with the same actions in different directions, and (2) learning an encoder that enforces invariance between these rotated trajectories with a Siamese framework. We show that our approach outperforms existing RL algorithms on 3-D locomotion agents like Ant, Humanoid and Minitaur.