Image-Based Backlash Width Identification for Eye-in-Hand Robotized Flexible Endoscopes

Robotic flexible endoscopes allow for more dexterous and ergonomic control than their manual counterparts. However, they are usually actuated by tendon-sheath mechanisms that exhibit backlash-like hysteretic behavior when operated. Various methods have been proposed to perform the model identification of these robots for control purposes, including the compensation of mechanical backlash, but they usually consider that the robotic system is visible to an endoscopic camera, which is not the case in robotic flexible endoscopes. In this paper, we introduce a novel method for automatic in-situ backlash identification in an eye-in-hand configuration, relying on embedded image measurements only. We propose to track scene landmarks in the image in order to estimate simultaneously their 3D positions and the parameters of the backlash model. The resulting learning problem is solved using a discontinuous extended Kalman filter to cope with the non-smooth model. The proposed approach is studied through simulations and then validated experimentally on a robotized endoscopic platform. The results show that the method allows to estimate non-constant backlash using only the standard embedded endoscopic camera and is thus applicable on-site for most flexible endoscopic robotic systems.