Robot programming by human demonstration: adaptation and inconsistency in constrained motion

Robot programming by human demonstration is an intuitive programming method designed to facilitate short term robotic applications. The programmer demonstrates the task using a teaching gripper that measures the human's forces and positions, and a robot program is generated from the demonstration data. A direct approach to generating a robot program would be to simply duplicate the demonstrated trajectory, yet with this approach the robot would not be able to adapt to the environment and would also duplicate unnecessary motion included in the demonstration. This article presents an approach for identifying a range of acceptable robot force and motion trajectories from multiple human demonstrations. The resulting robot program can adapt to part misalignment in simple assembly tasks. In addition, inconsistent human motion is not duplicated by the robot, but rather is used to identify task accuracy requirements. The analysis is applied to simple assembly tasks consisting of 3D translation. For each segment of constrained motion, a range of robot compliance is identified that can adapt to part misalignment, and limit robot errors to the task accuracy requirements.