Kyeonah Yu

August 1995

By definition, the clearance between fixtures and the parts they hold is small. When loading is performed by a robot, uncertainty in the initial position and orientation of the part and uncertainty in robot velocity as the part is moved make it a challenge to reliably and repeatedly load parts into fixtures. The ``3-2-1'' approach to fixture loading is well-known in industry: load the part into 3-point contact with a reference surface, slide it into 2-point contact with the fixture, then translate it along the pair into 3-point contact, and finally apply a clamp. This method implicitly assumes force feedback (since different part surfaces are emphasized as contacts are detected) and has not been geometrically formalized. In this thesis we present a planning algorithm that generates loading plans in the plane for sensor-based fixtures using the theory of Compliant Motion Planning. As an alternative to force sensing, we use simple binary sensors at each fixel to signal contact; the fixture state is represented by a binary vector. We extend existing theory to generalized polygonal environments and treat uncertainty in 2 phases. Phase 1 eliminates uncertainty in position and velocity using purely translational compliant motions. Phase 2 eliminates uncertainty in part orientation using a combination of rotational and translational compliant motions. We describe a selective compliance mechanism (SCM) that permits separate treatment of these phases. The result are conditional loading plans that branch based on fixture state. Finally we propose an alternative approach to eliminate uncertainty in part orientation by introducing {\it aligning pins} that are used to uniquely orient the part during loading. Once oriented, the SCM is locked to maintain part orientation during additional translations.