Assembly with robots involves two kinds of motions, those that are point-to-point and those that are force/torque guided, the former kind of motions being faster and more amenable to automatic planning and the latter kind being necessary for dealing with tight clearances. In this paper, we describe an assembly motion planning system that uses descriptions of assemblies and CAD models of parts to automatically figure out which motions should be point-to-point and which motions should be force/torque guided. Our planner uses graph search over a potential field representation of parts to calculate candidate assembly paths. Given the tolerances of the parts and other uncertainties, these paths are then analyzed for the likelihood of collisions. Those path segments that are prone to collisions are then marked for execution under force/torque control. The calculation of the various motions is facilitated by an object-oriented and feature-based assembly representation. A highlight of this representation is the manner in which tolerance information is taken into account: Representation of, say, a part contains a pointer to the boundary representation of the part in its most material condition form. As first defined by Requicha, the most material condition form of a geometric entity is obtained by expanding all the convexities and shrinking all the concavities by relevant tolerances. An integral part of the assembly motion planner is the execution unit. Residing in this unit is knowledge of the different types of automatic EDR (error detection and recovery) strategies. Therefore, during the execution of the force/torque guided motion, this unit invokes the EDR strategies appropriate to the geometric constraints relevant to the motion. This system, called AMP-CAD, has been experimentally verified using a Cincinnati Milacron T3-726 robot and a Puma 762 robot on a variety of assemblies.
Date of this Version