Integration of task scheduling, sensing, planning, and control in a robotic manufacturing work-cell

This dissertation is devoted to one of the most challenging research issues, the integration of low level sensor data and simple control mechanisms with high level perception and behavior to achieve a flexible and intelligent manufacturing system. Graph theory and Max-Plus algebra are applied as excellent mathematical tools to model and analyze the relationship among machine module's operations, called the task scheduling problem. The Max-Plus algebra model can be used for on-line task scheduling. A solution of the Max-Plus Model gives a task reference to be followed by all the operations in the manufacturing system.;A centralized multi-sensor data scheme provides information needed in real-time task scheduling and planning in the manufacturing system. Meanwhile, a novel calibration-free stereo vision algorithm is developed by using the shape information of a disc conveyor. The vision algorithm can localize targets in a moving coordinate frame, and is capable of making the tedious and time-consuming process of stereo visual sensor calibration no longer necessary. The ability to reconfigure the work-cell is thus improved. Moreover, the application of the proposed vision algorithm can be easily extended to general manufacturing systems by adding circular-shaped marks in the horizontal planes.;The theory of event-based planning and control is extended from single segment task planning to include multiple segment task planning. From the solution of the Max-Plus Algebra model, a unified action reference for multiple segment tasks can be found, through which tasks and actions in the manufacturing system are synchronized. At the same time, the manufacturing system gains the ability to deal with both discrete and continuous unexpected events and uncertainties. Most important, the unified action reference can consolidate discrete commands with continuous commands, such that a novel framework is established to integrate high level task scheduling with low level sensing, planning and control. Finally, all the theoretical results above are tested by a set of experiments in an automated robotic manufacturing work-cell. The experimental results demonstrate the feasibility of the theory in the laboratory environment.