Ergodic Graph Exploration via Markov Chain for Active Robotic Information Acquisition

Abstract

Many robotic applications can be considered as information acquisition, including surveillance,environmental monitoring, disaster response, and robotic learning. These tasks are often a combination of being repetitive, time consuming, time sensitive, or dangerous, which are unsuitable for human to perform. Specifically, this work consider inspection of confined spaces as the primary example. This work presents a semi-autonomous robotic system for assisting human to perform inspection in such hazardous environment. Challenges arise for robots to operate autonomously in these cluttered, poorly illuminated environment with complex connectivity such as localization, mapping, and navigation. Teleoperation also poses challenge as communication is limited with the confined space being enclosed in large metallic structures. This work focus on autonomous inspection with regard to foreign object debris (FOD) detection. First, an statistical FOD detection method is presented, accounting for mapping uncertainty in various location of the tank. The result is verified by the operator at the end of each inspection session. Second a hierarchical planning method is presented for optimizing the detection rate of FOD while handling the complex connectivity and limited navigation capability. Last, the planning method is generalized to a multi-robot system for collecting information in a large complex environment.