Maria T. Tagliaferri and Inseung Kang have trained a balance controller for an ankle exoskeleton on a simulated human, according to a paper on arXiv. A muscle-driven body model, built from one person's motion-capture data, acted as a virtual human for reinforcement learning of the controller.

The controller assists both ankles using only inertial sensors, without detecting the disturbance itself. Tested on the exoskeleton with five healthy people whose data were not used in training, it raised the backward floor movement they could take without stepping by 11.8%. The benefit varied with each person's balance strategy and body mass.

Source: arXiv Robotics (cs.RO)

Background

Maria T. Tagliaferri

Robotics researcher

Robotics researcher.