Integrating sensing, task planning and execution for robotic assembly
Abstract
Planning, sensing, and task execution are all essential parts of an intelligent robotic assembly system. In this dissertation, we have developed a coherent framework in which planning, sensing and task execution are intertwined and where each can benefit from the capabilities embedded in the other two. The resulting system is called IRAS for Integrated Robotic Assembly System. At the heart of IRAS is a geometric learning module and a planning module. The geometric learning module automatically deduces high-level descriptions of assembly goals by monitoring a human instructor as he manipulates objects. These assembly goals are then transformed into robot manipulation programs by the planning module, which consists of a nonlinear constraint-posting planner coupled with an opportunistic assembly execution planner. The nonlinear planner gives the system great efficiencies in searching through alternative action and sensing sequences; the execution planner allows different possible completions of partially executed plans, depending on the parts and other resources that are currently available. In order to allow the system to use more flexible grippers, we have also developed a theoretical framework for generating optimal two-fingered force-closure grasps. The implemented IRAS system has been used to successfully learn and execute various assembly tasks that involve parts located in random positions and orientations in a robotic workcell.
Degree
Ph.D.
Advisors
Kak, Purdue University.
Subject Area
Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.