Modeling and control of heterogeneous field robots under partial observation

Abstract

The control of large-scale dynamic systems, such as field robots, requires a more modular and scalable approach than traditional control theory. Recently, we designed modular supervisory controllers for field robots in agricultural applications based on a hybrid system that combines continuous-time and discrete-event systems. We verified that the developed hybrid system could control field robots and observe all the events while meeting the behavioral specifications. In contrast, all possible events cannot be monitored by a supervisor in a real-field robot system because of the presence of sensors, noise, disturbances, and failure. Therefore, in this study, we expanded the modular supervisor presented in previous studies by considering partial observations. Specifically, we proposed a process for designing an appropriate modular supervisor by considering the observability. The experimental results demonstrated that only observable events could cause state changes, verified by dynamic simulations representing a natural field environment. The effects of the proposed hybrid system-based modeling and control methodology are discussed in light of our systematic results.