Asymmetric Braided Artificial Muscles with Precise Electrothermal Actuation Control Enabled by Deep Learning

Abstract

Liquid crystal elastomers show promise for artificial muscles, but challenges remain in achieving excellent actuation performance and controllability under diverse operational conditions. This study presents a novel asymmetric braiding method using a Maypole braiding machine to integrate carbon nanotube yarns with liquid crystal elastomer fibers, producing an electrothermal fiber-shaped actuator. The actuator demonstrates exceptional performance in both air and water. In air, the actuator lifts 261 times its own weight (0.17 MPa) within 2.5 s, achieving a 45% contraction with a strain rate of 18%·s–1. Underwater, it reaches a 32% contraction within 3 s. To enhance controllability under diverse conditions, a long short-term memory (LSTM) model was proposed and applied, accurately predicting actuation strain with a coefficient of determination (R2) of 0.994. Applications in a music robot and underwater claw highlight its potential for flexible robotics, validating its advantages in programmable control, rapid response, and adaptability across environments.