A decentralized multi-agent control approach for robust robot plan execution

Abstract

Autonomous robots are evolving with versatile sensors and actuators that cooperate to allow the robot higher autonomy and more powerful capabilities. The complexity and difficulty of controlling such a heterogeneous robotic system have greatly increased, especially when the robot is intended to serve in a dynamic and open environment. When operating in non-structured, real-world environments, the robot plan execution may easily fail due to unexpected run-time contingencies. To robustly execute the plan, the robot system aims to keep sensing the plan execution status and to adapt the plan towards possible execution contingencies. Therefore, the issue of effectively coordinating sensing and acting (plan execution) behaviours becomes extremely challenging for complex robotic systems. The multi-agent approach presents a decentralized way to control the cooperative robots in a team and demonstrates scalability and flexibility while maintaining the robustness and modularity of robot organizations. The decentralized method of modelling and controlling the robot teams provides an approach for handling the complexity of controlling distributed and heterogeneous robotic systems. This article proposes the application of the multi-agent approach for the decentralized control of a single robotic system, which abstracts diverse robot components as different roles of heterogeneous agent entities. The multi-agent system allows flexible and efficient coordination between the robot sensors and actuators, which increases the run-time feedback for plan execution monitoring. Furthermore, to demonstrate the efficiency and effectiveness of the decentralized control approach, we have implemented the motivating example into a multi-agent system and conducted a comparative case study in a laboratory environment.