A theoretical framework for locomotor learning

Humans are usually able to effectively adapt their walking gait to novel situations such as diverse terrain, novel exoskeletons and prostheses, walking on split-belt treadmills, and other modifications to the person or the environment. Here, we provide a unified theoretical account of many qualitative features found in locomotor learning or adaptation, across multiple adaptation tasks. Our mathematical model posits that at the short time-scale humans respond via the feedback controller that maintains stable locomotion (that we had previously characterized) and at longer time-scales, humans slowly change this controller in a manner than reduces energy, following a negative gradient estimated from exploratory step to step variability (as in reinforcement learning). Using this framework, we predict changes in symmetry, entrainment, and energy expenditure in multiple tasks including walking on a split-belt treadmill and with exoskeletons. We also comment on the role of learning rates, sensory and motor noise, exploratory variability, the steepness and curvature of the energy landscape, etc., in effective locomotor learning.

Dr. Nidhi Seethapathi