Abstract
Multi-agent collaboration and coordination have been studied across various domains, yet challenges persist, particularly when applied to real-world dynamic environments. These challenges are exacerbated when dealing with physical agents, or robots, which face additional constraints such as limited moving speed and communication ranges. Moreover, the complexity increases with the deployment of heterogeneous agent teams, each possessing distinct capabilities, necessitating a balance between independence and adaptability in task scheduling based on the immediate presence of other agents. To address these challenges, a general multi-agent collaboration and coordination toolkit is designed and developed in this work. Our approach significantly enhances multi-agent collaboration and coordination, optimizing resource utilization, minimizing downtime, and boosting task execution efficiency. This toolkit allows agents to generate collaboration requests for task execution, aimed at minimizing the overall completion time of a set of given tasks. These requests are evaluated based on the receiving agent’s current task queue, along with the time and cost necessary to undertake the task and potential task execution quality. After receiving responses from other agents, the requesting agent selects the most suitable candidate based on a comprehensive assessment of total effectiveness and the potential for the earliest completion. Consequently, tasks are allocated to the most efficient agents, ensuring optimal task distribution and execution within the network. This collaboration and coordination toolkit is designed based on Taems. This domain-independent task modeling language allows the description of complicated relationships among tasks and sub-tasks and multiple-dimension criteria of task execution. This toolkit is tested with collaborative virtual robots developed on an innovative platform, which simulates a physical world environment and supports modeling of each robot’s sensory and actuator capabilities. This research aims to study and refine multi-agent communication and collaboration within simulated environments, focusing on task assignment and scheduling to elevate overall task performance and reduce periods of inactivity. A proof-of-concept demonstration presented within this thesis showcases the effectiveness of our methodologies in achieving these objectives.