Facilitating Cooperation in Multi-agent Robotic Systems
This thesis presents a new framework for developing, coordinating, and managing a system of situated agents for operation in distributed spatial environments. The framework was developed using agent-oriented software engineering techniques that consider concurrent features of real systems. It provides a realistic development environment that ensures reliable deployment of coordination schemes and mechanisms in real contexts. The framework uses multi-threaded agents so all the agents are executed concurrently, and each has control over its own behaviour. Such a model allows us to capture the behaviour of agents under realistic conditions where all the agents act concurrently. In addition, the framework employs Coloured Petri Net technique to model multiple, concurrent conversations in which a given agent may be engaged. In this framework agents must resolve complex and interdependent tasks. A Coloured Petri Net modelling technique was used to specify task plans and for the support of mechanisms both for the agents to identify task requirements and for managing the order in which various subtasks are performed. Our framework also enables the developed agents to identify subtasks that must be executed concurrently. The framework takes into account the unreliability of agent communication in real, physical environments by employing a timeout mechanism to test situations when communication responses are lost or delayed. It uses an environmental system time to accommodate the time delay associated with agents' responses in a distributed multi-agent system. Any agent framework must support the ability of agents to adapt according to the new circumstances. Our framework enables the agents to adapt in various ways. Agents can interact for solving various problems by changing their roles. Agents can change their roles at run-time. In addition, certain strategies were designed for agents to show how they can change their cooperation strategy in different situations to maximise their reward. A reinforcement learning mechanism was used to enable the agents to adopt the best strategies under changing circumstances. The framework that we have developed uses the Opal agent platform which complies with standards provided by Foundation for Intelligent Physical Agent (FIPA). Agents in the framework use asynchronous communication, and agents' messages are sent over a network. In addition, all the messages in the system use FIPA agent communication languages. By providing a design environment that enables the direct deployment of developed methods on real, instantiated agent robot systems, this work enhances the robustness of the agent-oriented software engineering process for mobile robots. A cooperative task execution scenario was tested using commercially available mobile robots (Garcia). The experiment demonstrated the efficiency and effectiveness of our framework in developing and deploying robots organisational models in physically distributed scenarios.
Advisor: Purvis, Maryam; Purvis, Martin
Degree Name: Doctor of Philosophy
Degree Discipline: Information Science
Publisher: University of Otago
Keywords: multi-agent systems; Coloured Petri Nets; robotics
Research Type: Thesis