Extended Kalman Filter Sensor Fusion Signals of Nonlinear Dynamic Systems.

摘要

World modeling for achieving operational space motion control of robot arms requires accurate measurements of positions and velocities in both joint and operational space. Servomotors used for joint actuation are normally equipped with position sensors and, optionally, with velocity sensors for interlink motion measurements. Further improvements in measurement accuracy can be obtained by equipping the robot arm with accelerometers for absolute acceleration measurement. In this report an Extended Kalman Filter is used for multi-sensor fusion. The real-time control algorithm was previously based on the assumption of a jerk represented as a white noise process with zero mean. In reality, the accelerations are varying in time during the arm motion and the zero mean assumption is not valid, particularly during periods of fast acceleration. In this report, a model predictive control approach is used for predetermining next-time-step jerk such that the remaining term can be modeled as Gaussian white noise. Experimental results illustrate the effectiveness of the proposed sensor fusion approach.