Manipulation Planning for Forceful Human-Robot-Collaboration

This thesis addresses the problem of manipulation planning for forceful human-robot collaboration. Particularly, the focus is on the scenario where a human applies a sequence of changing external forces through forceful operations (e.g. cutting a circular piece off a board) on an object that is grasped by a cooperative robot. We present a range of planners that 1) enable the robot to stabilize and position the object under the human applied forces by exploiting supports from both the object-robot and object-environment contacts; 2) improve task efficiency by minimizing the need of configuration and grasp changes required by the changing external forces; 3) improve human comfort during the forceful interaction by optimizing the defined comfort criteria. We first focus on the instance of using only robotic grasps, where the robot is supposed to grasp/regrasp the object multiple times to keep it stable under the changing external forces. We introduce a planner that can generate an efficient manipulation plan by intelligently deciding when the robot should change its grasp on the object as the human applies the forces, and choosing subsequent grasps such that they minimize the number of regrasps required in the long-term. The planner searches for such an efficient plan by first finding a minimal sequence of grasp configurations that are able to keep the object stable under the changing forces, and then generating connecting trajectories to switch between the planned configurations, i.e. planning regrasps. We perform the search for such a grasp (configuration) sequence by sampling stable configurations for the external forces, building an operation graph using these stable configurations and then searching the operation graph to minimize the number of regrasps. We solve the problem of bimanual regrasp planning under the assumption of no support surface, enabling the robot to regrasp an object in the air by finding intermediate configurations at which both the bimanual and unimanual grasps can hold the object stable under gravity. We present a variety of experiments to show the performance of our planner, particularly in minimizing the number of regrasps for forceful manipulation tasks and planning stable regrasps. We then explore the problem of using both the object-environment contacts and object-robot contacts, which enlarges the set of stable configurations and thus boosts the robot’s capability in stabilizing the object under external forces. We present a planner that can intelligently exploit the environment’s and robot’s stabilization capabilities within a unified planning framework to search for a minimal number of stable contact configurations. A big computational bottleneck in this planner is due to the static stability analysis of a large number of candidate configurations. We introduce a containment relation between different contact configurations, to efficiently prune the stability checking process. We present a set of real-robot and simulated experiments illustrating the effectiveness of the proposed framework. We present a detailed analysis of the proposed containment relationship, particularly in improving the planning efficiency. We present a planning algorithm to further improve the cooperative robot behaviour concerning human comfort during the forceful human-robot interaction. Particularly, we are interested in empowering the robot with the capability of grasping and positioning the object not only to ensure the object stability against the human applied forces, but also to improve human experience and comfort during the interaction. We address human comfort as the muscular activation level required to apply a desired external force, together with the human spatial perception, i.e. the so-called peripersonal-space comfort during the interaction. We propose to maximize both comfort metrics to optimize the robot and object configuration such that the human can apply a forceful operation comfortably. We present a set of human-robot drilling and cutting experiments which verify the efficiency of the proposed metrics in improving the overall comfort and HRI experience, without compromising the force stability. In addition to the above planning work, we present a conic formulation to approximate the distribution of a forceful operation in the wrench space with a polyhedral cone, which enables the planner to efficiently assess the stability of a system configuration even in the presence of force uncertainties that are inherent in the human applied forceful operations. We also develop a graphical user interface, which human users can easily use to specify various forceful tasks, i.e. sequences of forceful operations on selected objects, in an interactive manner. The user interface ties in human task specification, on-demand manipulation planning and robot-assisted fabrication together. We present a set of human-robot experiments using the interface demonstrating the feasibility of our system. In short, in this thesis we present a series of planners for object manipulation under changing external forces. We show the object contacts with the robot and the environment enable the robot to manipulate an object under external forces, while making the most of the object contacts has the potential to eliminate redundant changes during manipulation, e.g. regrasp, and thus improve task efficiency and smoothness. We also show the necessity of optimizing human comfort in planning for forceful human-robot manipulation tasks. We believe the work presented here can be a key component in a human-robot collaboration framework.