Quantifying the uncertainties-induced errors in robot impact detection methods

In the context of human-robot collaboration, an efficient impact detection is essential for safe operation. Residual-based collision detection relies on the difference between the estimated and actual motor torques. However, in these model-based methods uncertainties affect the residual in the same structural way as a collision does, leading to potential false alarms. This paper proposes to quantify the influence of uncertainties on residual generation methods based on the inverse dynamic model for both rigid and elastic-joint robots. These uncertainties-induced errors are investigated depending on their origin (parameters estimation or numerical differentiation). Boundaries of these errors are determined along a given trajectory and account as the minimum threshold of detectability of a collision. These results are illustrated in simulation.