Metareasoning to Adapt Path Planning Decision Making for Different Modalities of Uncertainty

Abstract Determining the appropriate level and composition of reasoning for decision-making is a critical component of variable autonomy, as an autonomous system might shift the responsibility for decision-making to its human teammate when it cannot solve the problem that has appeared. Metareasoning (reasoning about reasoning) can provide the autonomous system with a way to adapt its own reasoning (path planning) processes so that it relies less upon its human teammate. A use case was selected to study metareasoning for the problem of path planning in the presence of different types of uncertainty. We created and evaluated the performance of metareasoning policies that changed the path-planning approach for a set of simulated autonomous systems for diverse and dynamic uncertainties. The initial simulation included a single vehicle navigating from a start location to a specified goal while avoiding obstacles. Various iterations of the simulation were conducted, including missions with no uncertainty, uncertainty of obstacle locations, and uncertainty of obstacle status. Our simulation results showed that the vehicle used shorter paths when employing metareasoning, which also reduced total computational effort. These results suggest that metareasoning is a feasible strategy for variable autonomy human-robot teaming and that autonomous systems can use metareasoning to reduce the dependency on human intervention.