Robotic collision detection based on a speed-adaptive nonlinear gain momentum observer

Abstract

Robots have found widespread applications across industrial and service domains, making reliable collision detection increasingly important for safe human–robot collaboration. This paper presents a collision-detection method based on a Speed-Adaptive Nonlinear Gain Momentum Observer (SNGMO). The proposed framework integrates an enhanced Stribeck friction model, velocity-dependent observer gain scheduling, and a time-varying threshold to address dynamic-model uncertainty, low-speed friction disturbances, and the trade-off between residual suppression and collision response. First, the Stribeck model incorporates load-dependent and wear-related parameter variations to represent friction changes associated with operating load and equivalent wear state. Second, the SNGMO adjusts its natural frequency, gain, and damping according to joint velocity, increasing bandwidth at higher speeds for faster disturbance tracking while reducing bandwidth and increasing damping at low speeds and during direction reversals to suppress residual fluctuations and measurement noise. Finally, the threshold is adjusted online using motion-state and model-uncertainty information to distinguish external collisions from internal residual variations. The wear-related component was evaluated through representative parameter-drift simulations, and the complete method was experimentally validated on a UR5e robot. Under the evaluated conditions, the SNGMO achieved mean detection delays of 26.4–28.7 ms and reduced the average joint-wise collision-free residual root mean square by approximately 53% relative to the conventional GMO. These results indicate improved residual suppression and collision-response performance under the tested conditions.