If a robot has to perform a specified manipulation task involving intentional environmental contacts, a certain response behavior is desired to reduce strains and ensure successful completion without damage of the contacting bodies. On the other hand, the dynamic behavior of a manipulator depends strongly on its position and the gains of its joint controllers. Hence, varying these parameters for an optimized performance during manipulation seems to be an obvious task. In order to deal with impacts, oscillations and constrained motion, a model-based optimization approach is suggested, which relies on a detailled dynamic model of the manipulator incorporating finite gear stiffnesses and damping. These models are used to define an optimization problem, which is then solved using numerical programming methods. It is illustrated with an assembly task, namely inserting a rigid peg into a hole with a PUMA 562 manipulator. The expected advantage in industrial applications is a comparatively easy implementation, because performance can be improved by simply adjusting 'external' parameters as mating position and coefficients of the standard joint controller. Particularly no modifications of the control architecture and no additional hardware are required. Application of the proposed approach to a rigid peg-in-hole insertion under practical constraints can reduce the measure for impact sensitivity by 17%, that for mating tolerances by 78% and the damping of end-effector oscillations and motor torques by up to 79%. These improvements are shown to be reproducable experimentally.