Payload estimation for a robotic system using unsupervised classification

Abstract

A robotic system may be affected by external disturbances and parametric uncertainness, which change its dynamical behavior. One of the most common disturbances is the payload variation that affects the control system performance. If the payload variation is known, its negative effects can be minimized adjusting the control system parameters. However, when the payload variation is unknown, the control system parameters cannot be adjusted appropriately. This paper proposes a methodology for the payload variation estimation for a robotic system using unsupervised classification techniques. BSAS, MBSAS and Kmeans algorithms were employed as clustering techniques. The Silhouette index and the standard deviation were employed as performance indexes to compare the classification algorithms. Results showed that Kmeans algorithm has a better performance for the payload variation classification.