Stratégies de commandes stables des robots manipulateurs industriels à architecture de commande interne inconnue et inaccessible, évoluant dans un environnement déterministe ou aléatoire

Abstract

Nowadays, industrial robots realize more complex tasks, such as human-robot collabo ration and the flexible prehension of objects in unstructured environments where they are subject to random disturbances. In such environments, effective control requires the robot to interact with its environment, and therefore the development of torque control stra tegies. However, torque control is not always possible, because for reasons of safety and protection of intellectual property, many manufacturers produce industrial robots with an unknown and inaccessible internal control architecture. In this context, task space control is the most efficient. This control strategy is regularly confronted with the problem of glo bal stability. This problem of stability is accentuated when the architecture of the internal controller is unknown and inaccessible, and even more so when the robot is evolving in a random environment. In this thesis, we propose firstly an adaptive external control ler based on a radial function neural network, which approximates the dynamics of the unknown and inaccessible internal controller in order to impose the dynamics desired by the user by eliminating the effects of the internal controller in the control loop. We then propose a hybrid adaptive control approach that combines an indirect adaptive method for rejecting deterministic disturbances and a direct method for rejecting random distur bances. The use of Lyapunov theory enables us to demonstrate that the proposed control laws ensure semi-global closed-loop stability. The simulations carried out give trajectory following performances of the order of 1 × 10−5 m in a deterministic environment and of the order of 1.5 × 10−4m in a random environment. The results of experiments carried out on Intelek’s SCORBOT-ER and Cobot’s UR5 robots gave trajectory-following per formances of the order of 1 × 10−4m in a deterministic environment and 1.8 ×10−3m in a random environment.