Reproduction of stepping maneuver to prevent falling with a computational model.

Abstract

Human seems to maintain the upright posture by controlling and combining the hip, knee and ankle joints adequately. Especially when the posture is disturbed much mechanically, human steps the leg forward and resists the disturbance to prevent falling. The adaptive maneuver to step forward is observed quite often in a daily life, but the stepping mechanism is not elucidated enough from the neural and mechanical point of views. To understand the mechanism stepping maneuver, we developed a rigid-body link model to emulate the human stepping maneuver. In this study, our model is consisted of four rigid bodies and the model moves in two–dimensionally. The lower extremities have the shank and thigh including the foot, and the distal point of the leg on the stepping side is free end, while the distal point of the support side is placed on the floor and rotates as the ankle motion. The model learned the stepping maneuver with reinforcement learning. In this study, the reward function was expressed as standing the upper body, minimizing the joint torques, and maintaining the posture in a long time possibly. In the early learning, the model didn’t stand. However, after learning enough, the model kept standing by stepping forward even when a disturbance of load was applied the body. Even though our model isn’t driven by the muscle forces or not enough similar to the human musculoskeletal structure, the model could be useful to understand the mechanism of the stepping maneuver.