A dynamic planning system for automated manufacturing environments

This doctoral thesis presents a distributed planning system designed to support the characteristics of a dynamic, uncertain environment. To meet performance requirements, the planning system forms a parallelized plan which would be directly distributed into different groups of agents acting mostly independently, with only minor overlaps in common resources. The time requirements are critical in an automated manufacturing environment, therefore the selected plan should have minimal requirements for execution time. A planning framework for a dynamic, uncertain environment would also allow actions to have different effects for complete or failed execution, to provide mechanisms for actions recovery and to incorporate failure avoidance mechanisms. The known deterministic features of the system are captured into structures, that form the precompiled knowledge which is used by the planning algorithm. The algorithm isolates parts of the initial goal that can be achieved by different groups of agents. If the distribution does not seem feasible, a global planning is required. Otherwise, planning is attempted within the assigned groups of agents. A sensor-based monitoring mechanism detects possible errors. A recovery mechanism takes care of unsuccessful actions and unpredictable events such as errors and failures. All actions that are dependent on the unsuccessful actions will be also marked unsuccessful; so that these actions will be re-initiated.