Abstract
The control of tendon-driven robotic fingers presents significant challenges due to their inherent underactuation, coupled with complex non-linear dynamics arising from tendon elasticity, friction, and external disturbances. Therefore, achieving precise control of finger motion and contact interactions necessitates advanced modeling, estimation, and control strategies capable of addressing uncertainties in tendon tension, routing, and elasticity. This paper presents a comprehensive comparative study of three distinct control paradigms: feedback linearization with Proportional-Derivative (FBL-PD) control, feedback linearization with super-twisting sliding-mode algorithm (FBL-STA), and deep-deterministic reinforcement learning (DDPG-RL), for the precise trajectory tracking of a three-link tendon-driven robotic finger. Through extensive simulations, the performance of each controller is rigorously evaluated based on trajectory-tracking accuracy and robustness to varying disturbances. The results indicate that under disturbance-free conditions, the FBL-PD and FBL-STA controllers, when properly tuned, achieve precise tracking of the reference trajectory; however, they produce noticeably noisy control signals. When subjected to external disturbances, these controllers exhibit increased sensitivity, producing even noisier responses. In contrast, the DDPG-RL maintains smooth control dynamics and achieves sufficiently accurate tracking in both scenarios. This comparative analysis elucidates the strengths and weaknesses of each control strategy, offering critical insights and practical guidelines for the design and implementation of advanced control systems for dexterous tendon-driven robotic fingers.