Handover planning for every occasion

In this paper we explore shared manipulation tasks involving collaboration between two agents: a bimanual manipulator and a movable humanoid. We propose a deterministic planner that outputs a handover pose for the object manipulated, the grasps to be used for both agents, and arm trajectories that allow the agents to reach, grasp, interchange and place the object at its final goal pose. Most existing approaches to shared manipulation assume that both participating agents are (or can be) positioned face-to-face. In this paper, we consider a more general set of scenarios in which the relative pose between agents is not tightly constrained to facing each other. To accomplish a successful interaction between the agents, careful planning must be done in order to select the best possible handover pose such that the movable agent should not need to move its torso, only its arm, to interact with the object. We present 3 simulation experiments with manipulation tasks performed by a bimanual fixed manipulator and a humanoid robot.