Constrained Reinforcement Learning for Autonomous Farming: Challenges and Opportunities

Yongshuai Liu, Taeyeong Choi, Xin Liu

Abstract: Reinforcement learning (RL) is a machine learning technique to optimize policies to execute the most desirable sequence of actions in a complex environment. In particular, the optimal set of parameters is automatically learned based on observations while continuously interacting with environments. While this powerful approach has shown great success in various applications (e.g. games, robotics, etc.), relatively little attention has been paid to the agriculture domain. In this paper, we first discuss a general framework for RL with constraint terms for agricultural scenarios and also explore the potential challenges in developing it into a successful model under realistic environments. We then introduce potential data-driven strategies to effectively mitigate those challenges to realize fully autonomous systems for farm management.