Recursive algorithm for motion primitive estimation

The need for knowing future manipulator motion arises in several robotics applications, including notification or avoidance of imminent collisions and real-time optimization of velocity commands. This paper presents a real-time, low overhead algorithm for identification of future manipulator motions, based on measurements of prior motions and the instantaneous sensed actuator velocity commanded by an operator. Experimental results with a human-controlled, two degree-of-freedom manipulator demonstrate the ability to quickly learn and accurately estimate future manipulator motions.