Controlling Autonomous Robots: Metareasoning, Adjustable Autonomy, and Competence-Aware Systems

Giving a robot, autonomous vehicle, or unmanned system a degree of autonomy can reduce the operator’s workload, but determining the best level of autonomy is a challenging problem when the robot will operate in a dynamic environment. To avoid (and recover from) failures that can lead to unsafe actions and unsuccessful missions, an adjustable autonomy approach can change the level of autonomy during a mission. This is related to metareasoning, which is reasoning about reasoning, or deciding how to decide. Also related to these domains is competence-aware autonomy, which enables a robot to learn the optimal level of autonomy for a given action. These similar domains have some overlap but also key differences: metareasoning is more general, adjustable autonomy is a class of metareasoning approaches, and a competence-aware system uses an enhanced technique for adjustable autonomy. This paper reviews and compares these concepts to clarify these ideas for those who are considering using them to improve the performance of their autonomous robots.