Sliding mode based fault diagnosis with deep reinforcement learning add?ons for intrinsically redundant manipulators

This article presents a fault diagnosis control scheme for intrinsically redundant robot manipulators based on the combination of deep reinforcement learning (DRL) approach and battery sliding mode observers. The DRL plays role detecting isolating possible sensor faults, thus generating an alarm pin-pointing source. in turn allows to compensate faults independently from actuator ones. latter are therefore detected isolated by set observers driven input laws designed according optimal reaching algorithm. In order design apply such observers, global feedback linearization is performed, which transforms multi-input-multi-output (MIMO) nonlinear model into chain double integrators. proposal analyzed assessed realistic conditions using PyBullet environment 7 degrees-of-freedom (DOFs) Franka Emika Panda manipulator reproduced.