Off-Policy Deep Reinforcement Learning without Exploration
This addresses the challenge of learning from pre-collected data in real-world applications where exploration is constrained, offering a novel solution for batch reinforcement learning.
The paper tackles the problem of deep reinforcement learning from a fixed batch of data without further exploration, showing that standard off-policy algorithms fail due to extrapolation errors. It introduces batch-constrained reinforcement learning, which restricts actions to align with the data, and demonstrates its effectiveness in continuous control tasks.
Many practical applications of reinforcement learning constrain agents to learn from a fixed batch of data which has already been gathered, without offering further possibility for data collection. In this paper, we demonstrate that due to errors introduced by extrapolation, standard off-policy deep reinforcement learning algorithms, such as DQN and DDPG, are incapable of learning with data uncorrelated to the distribution under the current policy, making them ineffective for this fixed batch setting. We introduce a novel class of off-policy algorithms, batch-constrained reinforcement learning, which restricts the action space in order to force the agent towards behaving close to on-policy with respect to a subset of the given data. We present the first continuous control deep reinforcement learning algorithm which can learn effectively from arbitrary, fixed batch data, and empirically demonstrate the quality of its behavior in several tasks.