Search papers, labs, and topics across Lattice.
This paper explores the intersection of reinforcement learning and control theory, highlighting their shared origins in dynamic programming while addressing the methodological and cultural divides that have emerged. By introducing adaptive control and actor-critic reinforcement algorithms, the authors propose a novel integration of these paradigms to enhance data-driven decision-making in classical locomotion control tasks. The findings underscore the potential for cross-pollination between the two fields, offering insights that could bridge the existing gap and foster collaboration among researchers.
Bridging the gap between reinforcement learning and control theory could unlock new synergies in optimizing unknown dynamical systems.
Reinforcement learning and control theory are two adjacent scientific fields that focus on optimizing the controller of unknown dynamical systems using feedback. While both fields have common roots in dynamic programming, they have evolved with distinct methodologies, goals, and cultures. Despite decades of mutual influence, a significant gap persists between the two communities. This tutorial introduces adaptive control, actor-critic reinforcement algorithms, and a new way to combine these two paradigms for data-driven decision making on a classical locomotion control problem. Our aim is to provide a foundation for understanding the core differences between the two approaches and insights to help experts in each field better understand and engage with the tools and approaches of the other.
ETH