Monitoring Methods for Metareasoning Approaches for Recovering From Path Planning Failures on Mobile Ground Robots

Abstract The planning algorithms that autonomous systems use sometimes fail to generate effective solutions, which prevents mission success. Metareasoning, which is “thinking about thinking,” is one way to mitigate planning failures. To give an autonomous robot the ability to recover from these failures, we implemented a novel metareasoning approach in the ARL ground autonomy stack known as Phoenix. The metareasoning approach controls the global and local path planners in that stack by stopping and restarting the associated ROS nodes. In any metareasoning policy, there are methods of control and methods of monitoring. This study investigates the implementation of a meta-level that uses select monitoring methods that identify path planning failures for autonomous ground robots. The two monitoring methods proposed are the fixed radius methods, which are based on the mission goals, and the expected progress methods, which are based on the robot’s movement capabilities. In this study, we evaluated how using these methods affected navigation performance for the path planning algorithms that are in the ARL ground autonomy stack. The results of our simulation study showed that some policies helped the robot recover from path planning problems, and the robot could get closer to its final goal. In other cases, however, the metareasoning policies disrupted the robot’s navigation process and degraded performance. Additional research is needed to understand when metareasoning policies can improve performance.