Developments on Autonomous Task Coordination Using a Continuous Motivation State

Abstract

We consider the problem of imbuing a control system with high-level behaviors by decomposing those behaviors into low-level tasks. We use the recently developed Motivation Dynamics framework [42,44] to coordinate these tasks autonomously, using a continuous decision state. We make extensions of the system into tasks with recurrent behaviors, multiple agents, and continuous relaxation of hybrid systems. In these extensions several proof-of-concept examples are given, and analytical performance guarantees are provided. We then move on to adapt the Motivation Dynamics framework to optimally perform according to formal specifications. We consider Signal Temporal Logic (STL) and its quantitative semantics, the robustness metric, as a tool to accomplish this. Several candidate robustness metrics are considered, with the goal of selecting one with good performance in numerical optimization. Comparison of the metrics yields a clear best-performing variant, and future adaptation to the Motivation Dynamics framework is discussed.