Improved Performance Prediction for Twisted String Actuator Under Variable Load

Twisted String Actuator (TSA) is an innovative artificial muscle that converts the rotary motion of an electric motor into linear motion of a load. TSA exhibits great potential for soft robots and robotic exoskeletons due to its high transmission ratio, simple structure, intrinsic compliance, and low cost. However, the actual dynamic behavior of TSA is greatly influenced by varying load conditions, making it challenging to accurately predict its behavior under different loads. This study addresses this issue by proposing a variable load model for TSA based on the backpropagation (BP) neural network. By training the neural network using extensive experimental data, we establish and validate the model against various load conditions. The proposed model outperforms conventional variable load mathematical models, offering smaller prediction errors and more accurate descriptions of TSA behavior under different load conditions. This finding has important implications for system design and control of TSA.