Information-Based Patrol Speed Control Method for Rail-Guided Robot System Using Deep Deterministic Policy Gradient Algorithm

Abstract

To manage the safety of multi-use facilities, many CCTVs and alarm sensors are used, however, they cannot replace patrol tasks to check site conditions from multiple directions. The developed rail-guided smart patrol robot helps alleviate the workload of managers by capturing images and measuring sensors at a desired location at a scheduled time in a separate space from visitors or workers. This paper proposes an adaptive patrol speed control algorithm to improve patrol performance in the facility environment. By applying the Deep Deterministic Policy Gradient (DDPG)-based learning model, the smart patrol robot can be allowed to move at an optimal speed according to the congestion of images captured in the field. The designed model can be trained by defining the reward function based on the entropy to maintain the obtained information. The proposed algorithm demonstrated performance in controlling patrol speed according to situation changes in a virtual multi-use facility environment.