Inference of analytical functions modeling human finger movement using symbolic regression
Question of interest
Can we infer functionally accurate mathematical regression functions modeling human finger movement directly from data?
A paper that I published that forms a prequel to this work can be found here.
Control of finger movement and force is made possible by a complex network of tendons.
Analytical functions modeling tendon excursions are useful for building robotic and prosthetic hands and understanding control of finger movement.
Hypothesis: Simultaneous extraction of form and parameters of functions directly from experimental data will give us more functionally
accurate representations as opposed to assumption of form for the expression apriori.
Regression: Model inference
Symbolic regression as implemented in the software Eureqa was used to regress
mathematical functions of the form s = f(theta1, theta2, theta3 and theta4) for each tendon.
The functional accuracy of the models was compared to that of vanilla polynomial regression and Landsmeer models on cross-validation datasets.
Below is a comparison of prediction accuracy and number of parameters on cross-validation datasets: Symbolic regression vs. Polynomial vs. Landsmeer model