Optimistic Active Exploration of Dynamical Systems
This addresses the challenge of efficient exploration in reinforcement learning for broad applicability to novel tasks, though it appears incremental as it builds on existing probabilistic modeling and optimal control methods.
The paper tackles the problem of actively exploring unknown dynamical systems to estimate a globally accurate model for solving multiple downstream tasks in a zero-shot manner, resulting in the OPAX algorithm that shows theoretical sample complexity bounds and performs well in experiments compared to heuristic approaches.
Reinforcement learning algorithms commonly seek to optimize policies for solving one particular task. How should we explore an unknown dynamical system such that the estimated model globally approximates the dynamics and allows us to solve multiple downstream tasks in a zero-shot manner? In this paper, we address this challenge, by developing an algorithm -- OPAX -- for active exploration. OPAX uses well-calibrated probabilistic models to quantify the epistemic uncertainty about the unknown dynamics. It optimistically -- w.r.t. to plausible dynamics -- maximizes the information gain between the unknown dynamics and state observations. We show how the resulting optimization problem can be reduced to an optimal control problem that can be solved at each episode using standard approaches. We analyze our algorithm for general models, and, in the case of Gaussian process dynamics, we give a first-of-its-kind sample complexity bound and show that the epistemic uncertainty converges to zero. In our experiments, we compare OPAX with other heuristic active exploration approaches on several environments. Our experiments show that OPAX is not only theoretically sound but also performs well for zero-shot planning on novel downstream tasks.