Estimating skier’s load in cross-country sit-ski by generative simulation of poling motion using deep reinforcement learning

Abstract

The objective of this study was to estimate the skier’s load during poling in cross-country sit-ski as well as its dependency on a change in equipment by generative simulation using deep reinforcement learning. As an example of the change in equipment, pole length was targeted in this study. Three cases of different pole length were considered, and the joint motions for those three cases were virtually created by the generative simulation. The results show that the simulation model of poling motion was validated by comparing the simulated motion and force with experimental motion and force. From the generative simulation, it was found that natural and smooth poling motions were successfully obtained for all three cases. Regarding the joint torques, it was found that the lumbus torque in the poling period exhibited the flexing direction first, followed by the extending direction, while the shoulder and elbow torque exhibited extending direction throughout the poling period. Regarding the effect of the change in pole length, it was found that the flexion torque at the lumbus and the extension torque at the shoulder became larger for longer poles, although the extension torque at the elbow was reduced.