Recurrent Neural Modeling and Controller Design for a Lightweight Actuator Driven by Cable Transmission for a Robotic Arm

This paper presents the modeling and controller design based on recurrent neural networks for a lightweight actuator driven by Bowden cable transmission with a novel constant tension mechanism for a robotic arm. The aim is to minimize the mechanical backlash typically associated with such systems. The actuator enables the robotic arm to offload motor weight entirely to an external structure, significantly enhancing payload capacity and reducing the arms inertia. We detail the challenges inherent in Bowden-based transmissions, particularly the considerable backlash, and introduce a solution incorporating a tension-regulating mechanism to maintain consistent cable tension during operation. The system dynamics are modeled using Recurrent Neural Network to simulate its performance under varying conditions. To achieve precise motion control, a Recurrent neural network-based controller is developed and implemented, optimizing the joints ability to follow complex trajectories and reducing backlash. A series of simulations using MATLAB demonstrates the actuators capacity to accurately perform trajectory-tracking tasks with a minimum absolute mean error of 2.3°, showcasing the effectiveness of the constant tension mechanism and control strategy. The results highlight the potential of this system for applications requiring lightweight, highprecision robotic arms, particularly in surgical and rehabilitation scenarios.