Reliable real-time robot operation employing intelligent forward recovery

Modern computer-controlled robots typically fail at their tasks whenever they encounter an error, no matter how minor. The physical environment of a typical assembly robot is too unstructured to benefit from conventional software approaches to reliability. We present an approach which interfaces the real-time operation of the robot with an intelligent subsystem that performs error analysis and forward recovery if a failure occurs. Our approach involves a special representation of the robot's program that efficiently tracks the robot's operation in real time and is easy to modify to include automatically generated recovery procedures.