Multi-model observer for position estimation and object contact detection of a flexible robotic actuator

This paper proposes a multi-model based approach that estimates joint position and collision detection of a lightweight electric actuator dedicated to manipulation tasks. Taking advantage of an optimized back-drivable mechatronic design, the use of proprioceptive measures at the motor level enables the estimation of the unmeasured terminal position of the mechanical motor-to-joint transmission and force contact disturbance. An accurate model of the actuated mechanical flexible transmission is theoretically introduced to precisely describe its non-linear dynamic behavior, and is reformulated into a polytopic form. Then, a multi-model observer-based estimator is synthesized for object contact detection strategy and joint position estimation without additional position/force sensor. A root-clustering synthesis is performed through Linear Matrix Inequalities (LMIs) and is experimentally validated with a single degree-of-freedom manipulator.