Constraint-based haptic rendering of point data for teleoperated robot grasping

We present an efficient 6-DOF haptic algorithm for rendering interaction forces between a rigid proxy object and a set of unordered point data. We further explore the use of haptic feedback for remotely supervised robots performing grasping tasks. The robot captures the geometry of a remote environment (as a cloud of 3D points) at run-time using a depth camera or laser scanner. An operator then uses a haptic device to position a virtual model of the robot gripper (the haptic proxy), specifying a desired grasp pose to be executed by the robot. The haptic algorithm enforces a proxy pose that is non-colliding with the observable environment, and provides both force and torque feedback to the operator. Once the operator confirms the desired gripper pose, the robot computes a collisionfree arm trajectory and executes the specified grasp. We apply this method for grasping a wide range of objects, previously unseen by the robot, from highly cluttered scenes typical of human environments. Our user experiment (N=20) shows that people with no prior experience using the visualization system on which our interfaces are based are able to successfully grasp more objects with a haptic device providing force-feedback than with just a mouse.