Disentangling Controllable Object through Video Prediction Improves Visual Reinforcement Learning
arXiv:2002.09136
Abstract
In many vision-based reinforcement learning (RL) problems, the agent controls a movable object in its visual field, e.g., the player's avatar in video games and the robotic arm in visual grasping and manipulation. Leveraging action-conditioned video prediction, we propose an end-to-end learning framework to disentangle the controllable object from the observation signal. The disentangled representation is shown to be useful for RL as additional observation channels to the agent. Experiments on a set of Atari games with the popular Double DQN algorithm demonstrate improved sample efficiency and game performance (from 222.8% to 261.4% measured in normalized game scores, with prediction bonus reward).
arXiv commentAccepted to ICASSP 2020
References in corpus (6)
- Trust Region Policy Optimization
- Model-Based Reinforcement Learning for Atari
- Learning Latent Dynamics for Planning from Pixels
- Incentivizing Exploration In Reinforcement Learning With Deep Predictive Models
- Visual Foresight: Model-Based Deep Reinforcement Learning for Vision-Based Robotic Control
- Independently Controllable Features