One Policy is Enough: Parallel Exploration with a Single Policy is Near-Optimal for Reward-Free Reinforcement Learning
This provides a theoretical foundation for efficient parallel exploration in RL, with practical implications for reducing complexity in exploration phases, though it is incremental as it builds on existing reward-free RL frameworks.
The paper tackles the problem of parallel exploration in reward-free reinforcement learning for linear Markov decision processes and two-player zero-sum Markov games, showing that using a single policy across all agents achieves an almost-linear speedup compared to sequential methods and is near-minimax optimal.
Although parallelism has been extensively used in reinforcement learning (RL), the quantitative effects of parallel exploration are not well understood theoretically. We study the benefits of simple parallel exploration for reward-free RL in linear Markov decision processes (MDPs) and two-player zero-sum Markov games (MGs). In contrast to the existing literature, which focuses on approaches that encourage agents to explore a diverse set of policies, we show that using a single policy to guide exploration across all agents is sufficient to obtain an almost-linear speedup in all cases compared to their fully sequential counterpart. Furthermore, we demonstrate that this simple procedure is near-minimax optimal in the reward-free setting for linear MDPs. From a practical perspective, our paper shows that a single policy is sufficient and provably near-optimal for incorporating parallelism during the exploration phase.