| Περίληψη: | This study explores the detection of collisions on a collaborative robotic arm (cobot) under different Case Studies that correspond to distinct robotic tasks using statistical time-series methods. The collisions cause abrupt changes in the investigated torque and force signals acquired from the test rig, and two unsupervised and one supervised method are postulated for the effective and early detection of them. The postulated methods are three, namely the variance-based, the wavelet-based (supervised) and the RAR model-based, which are comparatively assessed with each other, as well as with two state-of-the-art methods, namely the residual-based and the FFT-based (supervised), in terms of True Positive Rate (TPR), False Positive Rate (FPR) and Detection Delay Time (DDT). The three case studies investigated are based on the measurements and the results of previous works that use the same KUKA LWR4+ robotic arm and they include either the autonomous motion of the robot in the first two Case Studies or the physical collaboration between the human and robot in the third Case Study. The collisions are of different nature along the different Case Studies, with varying magnitude and direction. The results indicate great TPR via all three postulated methods, that outperform the state-of-the-art ones, while all methods have zero FPR. Regarding detection time, each method has better performance in a different Case Study, with the RAR model-based being the fastest overall, followed by the wavelet-based and then the variance-based one.
|
|---|
