Robotic Pre-manipulation - Real-Time Polynomial Trajectory Control for Dynamic Object Interception with Minimum Jerk

Abstract

This paper presents a method for capturing a free-moving object in the presence of noise and uncertainty with respect to its estimated position and velocity. The approach is based on Hermite polynomials and involves matching the state-space parameters of the object and the end effector at the moment of contact. The method involves real-time re-planning of the robot trajectory whenever new estimates of the object's motion parameters are available. Continuity in position, velocity, and acceleration is preserved independently of the planning update rate and the resulting trajectories are characterized by low jerk. Compared to other methods that directly solve for higher-order polynomial coefficients, the proposed algorithm is computationally efficient and does not require a linear solver. Experimental results confirm the advantages of this method during real-time interception of a dynamically moving object with continuous velocity estimation and high-frequency re-planning.