Reliable Robot Missions in Dynamic Environments Using Agent-based Controllers

Controllers for real-world robots are constructed based on expectation models which cannot account for all unexpected events. If unexpected events occur, the robot might either stop working altogether or do something that violates its given mission specification. In this paper, we aim to provide a more principled mechanism for dealing with specifications when encountering the inherent uncertainty of the real-world. We utilise agent-based control of a robotic system wherein a rational agent, programmed using a Belief-Desire-Intention (BDI) agent programming language, acts as a high-level robot controller. This agent is aware of its mission specification and is able to recognise violations of this specification caused by unexpected events. On recognising such a violation, the agent can switch to a more suitable plan. However, if there are no suitable plans, the agent can also weaken its mission specification dynamically. The agent can output not only which parts of the mission specification have been achieved but which parts have been weakened and why.