Decentralized Grasp Coordination and Kinematic Control for Cooperative Manipulation

Multi-robot systems have shown potential over single robot systems in handling long, large and heavy objects. Manipulation with multiple robots increases task precision and decreases required load capabilities of individual robots while saving cost and space. However, the limitations of multi-robot systems in communication, sensing and knowledge sharing and constraints that occur while performing cooperative manipulation pose challenges for coordination and control. This thesis focuses on multi-robot grasp planning and control of collaborative manipulation. First, the thesis examines the question how to plan cooperative grasps given a known object for a group of robots that are decentralized and heterogeneous. Decentralized grasp planning approaches which account for incomplete embodiment knowledge and utilize imprecise local information from vision are presented. In addition, to tackle observation constraints, strictly decentralized approaches which only require role assignments are also proposed. Grasp decisions from the approaches are based on evaluation of traditional grasp quality measures under uncertainty. All approaches aim to maximize the quality of cooperative grasp under decentralization and heterogeneity. The approaches are further extended to include task specific information and their usefulness to particular tasks are studied. For experimental study, a complete system pipeline for decentralized grasp coordination is developed and physical metrics to evaluate the success of cooperative grasps are presented. The main contribution of this part is the first extensive investigation of grasp planning for decentralized multi-robot coordination. Second, given a cooperative manipulation task for heterogeneous multi-robot system, the thesis studies, how to safely coordinate the motions under joint limit constraints. A kinematic controller that employs relative Jacobian for coordination of motions and a control strategy based on task prioritization and smooth activation to ensure safe collaborative manipulation and smooth joint limit avoidance are presented. Behaviour of the controller under different redundancy configurations and applicability in practice for collaborative manipulation are demonstrated.