Force Sensitive Robotic End-Effector Using Embedded Fiber Optics and Deep Learning Characterization for Dexterous Remote Manipulation

Abstract

Many of the tasks that require a high level of autonomy in complex and dangerous situations are still done by human operators with a high risk of accidents. Although various remotely controlled robot systems have been proposed, the remote operation has limitations in performance and efficiency compared with on-site operations. This letter proposes the design of a new force and tactile sensing mechanism for a robotic end-effector suitable for deployment in harsh environments with integrated force sensing based on fiber optic sensors embedded in a simple and rugged structure. The proposed end-effector was able to detect the magnitude and location of the applied force accurately for high-performance tele-manipulation using hierarchical deep neural network (root mean square errors of 0.43 and 1.11 mm for estimating the contact location in the x-axis and the y-axis, respectively, and 1.16 N for estimating the magnitude of the contact force). Gaussian smoothing was used to support the performance, reducing the error levels by 25%. Also, learning feasibility was performed based on the auto-encoder. Using preliminary bilateral remote control experiments, we demonstrated the feasibility of the telemanipulation with dexterity.