Divide-and-Conquer Reinforcement Learning
This addresses the problem of inefficient learning in stochastic environments for reinforcement learning practitioners, offering a novel algorithmic improvement.
The paper tackles the challenge of high-variance gradient estimates in model-free reinforcement learning due to initial state variation by partitioning the state space into slices and optimizing an ensemble of policies, which are unified into a single policy. The result shows that this divide-and-conquer RL approach greatly outperforms conventional policy gradient methods on complex tasks like grasping, manipulation, and locomotion.
Standard model-free deep reinforcement learning (RL) algorithms sample a new initial state for each trial, allowing them to optimize policies that can perform well even in highly stochastic environments. However, problems that exhibit considerable initial state variation typically produce high-variance gradient estimates for model-free RL, making direct policy or value function optimization challenging. In this paper, we develop a novel algorithm that instead partitions the initial state space into "slices", and optimizes an ensemble of policies, each on a different slice. The ensemble is gradually unified into a single policy that can succeed on the whole state space. This approach, which we term divide-and-conquer RL, is able to solve complex tasks where conventional deep RL methods are ineffective. Our results show that divide-and-conquer RL greatly outperforms conventional policy gradient methods on challenging grasping, manipulation, and locomotion tasks, and exceeds the performance of a variety of prior methods. Videos of policies learned by our algorithm can be viewed at http://bit.ly/dnc-rl