Twisting and coiling actuators (TCAs) are lightweight artificial muscles that can produce large linear contractions while lifting heavy loads with low power. A TCA consists of two or more strings connected to a rotational motor and a load. When activated, the strings twist and coil, causing contraction. This study presents a data-driven framework for modeling TCA behavior using experimental data. Polynomial regression, Support Vector Regression (SVR), and symbolic regression were applied using the motor angle and payload weight as inputs. The models were evaluated under various loads and implemented in a neck-rehabilitation prototype. SVR showed the highest accuracy (RMSE: 6.17 mm upward, 4.98 mm downward); however, it lacks a closed-form expression, limiting its control utility. Polynomial regression modeled the twisting phase well but struggled with coiling (RMSE: 19.56 mm downward). Symbolic regression balanced accuracy and interpretability, yielding closed-form expressions (RMSE: 8.98 mm upward, 9.35 mm downward, R2 > 0.996, making it practical for real-time applications. The study also examined the effects of payload (300-750 g) and motor speed. Upward actuation was sensitive to payload changes, whereas the downward motion remained stable. Motor speed had a minimal effect on contraction in both directions. These results highlight the importance of including payload weight as a model input to improve performance under different conditions.
