Metareasoning for robots

Metareasoning is the process of reasoning about reasoning. It monitors and controls a robot’s reasoning processes. Metareasoning gives a robot or autonomous system the ability to reason about its own reasoning processes. Metareasoning for robots must cope with the real-time nature of robots and the robot’s limited computational resources.

Using metareasoning delays some design decisions about the reasoning system. No single algorithm is ideal for every situation; for instance, some algorithms that are robust are inefficient. Instead of committing to a single algorithm, we can implement a metareasoning approach that selects the best one in real time. There are a wide variety of metareasoning approaches that can be applied.

Due to this flexibility, metareasoning can improve performance, and it gives the system the ability to deal with unexpected situations, which can also make its operation safer. Metareasoning is meant to improve (ideally, optimize) the performance of any system that has limited computational resources because of size, weight, and power constraints.

Although researchers have implemented limited metareasoning approaches and demonstrated their effectiveness, the next steps are to develop, deploy, and test metareasoning approaches in systems at higher technology readiness levels.