Observer-Extended Direct Method for Collision Monitoring in Robot Manipulators Using Proprioception and IMU Sensing

In this letter a novel method for accurate and high-bandwidth real-time monitoring of robot collisions is presented. To the authors' knowledge this is the first time the so called direct method, which is mathematically the simplest and theoretically the ideal one, has been realized at practically relevant levels. For this, joint velocity and acceleration of serial chain robots are initially estimated using observer techniques that fuse joint position, Cartesian acceleration and angular velocity measurements. Consequently, this algorithm, which also extends our previous work in velocity and acceleration estimation, together with the available robot dynamics model are utilized to algebraically monitor external forces applied to the robot. Specifically, the proposed sensor fusion setup increases estimation bandwidth and decreases detection uncertainties compared to existing methods. Moreover, since neither inversion of large matrices nor their derivatives are required, our approach also shows increased numerical stability. Finally, the developed algorithm is evaluated based on a realistic simulation with the consideration of all parasitic effects and experimentally with a 7-DoF flexible joint robot.