Robustness issues in path planning and control

Neither traditional robot planning, nor the more recent behavior-based paradigm promises to scale to interesting problem solving tasks for complex sensorimotor systems operating in unstructured environments. It is unreasonable to require complete and precise world geometry. Paths constructed on this basis alone ignore physics and so violate the precision assumptions that they require. Moreover, flexible, goal-oriented behavior can not be designed as a fixed reflexive response to local stimulus. The author argues for a design specification that addresses correctness, incomplete information, and uncertainty. The author proposes that truly intelligent automation can not be completely designed a priori, but must be structured to produce safe and relevant motor alternatives, and must adapt to varying problem solving contexts.