Coordinated Manipulation Under Distributed Impedance Control

Distributed impedance is an efficient approach for controlling any kind of cooperating manipulation system. It combines the advantages of decentralized control solutions and the robustness of impedance control with force feedback. Distributed impedance was implemented on a dualarm system composed of two industrial manipulators. Experimental results were obtained for free and constrained manipulation tasks. 1. T h e d i s t r i b u t e d i m p e d a n c e a p p r o a c h Coordinated manipulation is a generic term used to designate the manipulation of a single body by multiple limbs. This is illustrated by two manipulators carrying an heavy, volumic and/or flexible object. Dextrous hands constitute an interesting sub-class of cooperative systems. They offer a high degree of flexibility compared to basic on-off grippers. They are particularly useful for manipulating fragile objects. An analysis of the control problem of coordinated manipulation systems shows that besides the control of the trajectory and forces experienced on the manipulated object, it is necessary to provide a control of the internal forces. Control of internal forces indeed is critical with respect to grasp stability in that it permits to avoid slippage or lost of the contact on the surface of the object. Most previous approaches for cooperative system control consist in global and centralized methods [1]. Object coordinates and internal efforts must be evaluated in line to be independently servoed. As a consequence, a specific hardware development is needed for data centralization. Conversely, local approaches consider each cooperative manipulator as an autonomous, independently controlled system. Practically, each manipulator's controller is used without any hardwaremodification in the multi-robot system. This provides flexibility and modularity. Furthermore, undesirable phenomenons (actuators limitation, non-colocated modes,...) that limit the performances from a particular manipulator, can be locally treated, without major influence on the whole system. A first possible approach consists in implementing pure joint control [2]. In this case, computation of the grasp and robot inverse kinematic models are required. This makes joint level approaches very sensitive to uncertainties in