Cooperative Behavior-Based Reinforcement Learning Control With Vehicles

This article addresses the cooperative control of behavior-based nonholonomic mobile vehicles (NMVs). Null-space-based behavioral control (NSBC) has limitations in the particle model and mission centralization. A novel underactuated reinforcement learning behavioral control (URLBC) approach is proposed in this article to address these limitations. An underactuated mission paradigm is prepared to replace the particle model and prevent nonholonomic constraint violations of the behavioral command. Subsequently, a distributed reinforcement learning mission supervisor is developed to determine behavioral priorities without any centralized unit, intelligently generating behavioral commands while achieving expected priority switching. Furthermore, an identifier–actor–critic behavioral controller is formulated to track behavioral commands without exact models, thus effectively reducing control costs and model dependencies. Simulation shows that URLBC has fewer unsatisfactory priority switching instances and lower cumulative control costs than existing NSBC methods. URLBC is implemented on JetAuto robots to validate its practicality.