Dynamic coordination of multi-robots by Bayesian modeling

Many of the tasks required of robotic agents can be done more effectively and efficiently if they are provided with a capability of making generalizations, drawing inferences and extracting patterns by observation. A generic framework for incorporating learning for the purposes of dynamic rescheduling in a multiple robotic system is proposed. The proposed method utilities a statistical model to make use of run-time observations to recover underlying dependencies in the task domain which is then used to make dynamically optimal plans. The mathematical implementation of the model and assessment issues are discussed, and the approach is illustrated by example of a foraging task.