Algorithms for manipulation planning with imperfect parts and inaccurate manipulators
12 / 2011 - 12 / 2015
Automated design of manufacturing processes today is where computer technology was in the 1960s: a patchwork of ad-hoc solutions lacking a rigorous scientific methodology. What is missing is a framework for the systematic design of automated manufacturing systems that manipulate (e.g. assemble, hold, sort, feed) industrial parts. To be reliable and inexpensive, such systems often use simple physical actions by manipulators that require modest sensing capabilities. These characteristics make automation amenable to systematic analysis and synthesis. Recent research in algorithmic automation confirms this statement, but a severe idealization has to be removed to make results applicable in practice. A major obstacle to practical application is that existing results assume perfectly-shaped parts and physical actions that are executed with infinite precision, while real parts are manufactured to tolerances and real manipulators are subject to errors. The research in this proposal tackles this severe idealization as its goal is to design algorithms for planning manipulation tasks that are guaranteed to work despite manipulator inaccuracy and part imperfection. Taking into account these variations asks for alternative models and analyses of physical actions, and for completely different approaches to plan synthesis. As part geometry and motion are crucial to the definition, modeling, and solution of the problems that we address, we will employ algorithmic techniques and insights from the confluence of computational geometry and path planning. These techniques, supplemented by physical part properties, offer a prospect of algorithms for the automated design of reliable low-cost solutions in automated manufacturing.