Direct Joint Space State Estimation in Robots With Multiple Elastic Joints

For robots with joint elasticity, discrepancies exist between the motor side and the load side (e.g., the link of the robotic joint). Thus, the load side (end-effector) performance can hardly be guaranteed with motor side measurements alone. In this paper, a computationally efficient load side state estimation scheme is proposed for the multi-joint robot with joint elasticity, which is equipped with motor encoders and a low-cost end-effector MEMS sensor such as a three-axial accelerometer. An optimization-based inverse differential kinematics algorithm is developed to obtain the load side joint state rough estimates. With these rough estimates, the estimation problem is decoupled into simple second-order kinematic Kalman filter for each joint to refine the joint position and velocity estimates. Maximum likelihood principle is utilized to estimate the fictitious noise covariances used in the Kalman filter. Both offline and online solutions are derived. The extensions to other sensor configurations are discussed as well. The effectiveness of the developed method is validated through the simulation and the experimental study on a 6-DOF industrial robot.