Efficient Robot Control with Motion Primitives and Mixed Reality

Abstract

We introduce a method that combines simplified input commands and head orientation tracking to improve human-robot interaction efficiency in collaborative robot tasks. These simplified commands are expressed by operators as 3D input gestures using a haptic input device where the available gesture set is mapped to the primary ‘Motion Primitives’ required to command the robot, e.g. grasp, transport, place and rotate. The head orientation tracking provided by a mixed reality headset allows the operator to designate the target object or desired position of the target object when executing motion primitives. This combined input strategy enables operators to focus on high-level decision-making while offloading otherwise continuous and repetitive tasks to the robot. To validate our approach, we conducted a controlled user study (n = 16) emulating a brick paving teleoperation task and compared our method using motion primitives to a baseline method directly mapping input motion to robot motion. Our method significantly reduced grasping attempts, trajectory length, active control time, active control ratio, and perceived workload in comparison to the direct control method.