Research on Collision Detection and Collision Reaction of Collaborative Robots.

The first-order model-based generalized momentum observer (GMO) is a low pass filter with high-frequency attenuation characteristics and detection delay. Therefore, it cannot detect short impacts (extremely short time interval) sensitively. According to the frequency distribution of modeling errors, collision signal, and measurement noise, we design three different second-order model-based GMOs, which improve the performance of collision detection effectively. Specifically, these GMOs include a second-order damped system (abbreviated as Damp), a second-order damped system with PD regulation (abbreviated as PD), and a band-pass filter (abbreviated as BPF). Typically, the collision reaction function can be realized by switching to the torque control mode with gravity compensation. However, the robot cannot continue to move along its original trajectories after switching from position to torque control, and cannot continue to perform subsequent tasks. To address this problem, we design a collision reaction strategy in position control mode by mapping the output of the observer to the increment of joint velocity, realizing a safe collision reaction.