Anthropodynamics: Inferring the Control System Humans Use While Walking and Running

在正常情况下，健康的成年人可以走路和跑步，几乎不可能摔倒。这种保证水平远远超过了最先进的有腿机器人和机器人假肢设备的性能。这个项目的目标是利用人类实验的数据来逆向工程人类运动的动力学和控制——也就是说，通过观察人类自然行走和对地形变化、突然推搡和其他干扰的反应来理解人类运动的控制规律——并将这种理解应用于更自然的假肢装置和更有能力的有腿机器人。该项目的结果还将有助于深入了解老年人和其他高危人群的运动障碍和其他平衡问题，并允许设计更有效的平衡改善设备。在这个项目中，控制人类运动的生理控制规律将根据自然的无扰动行走和跑步的实验，以及在运动过程中对精心选择的外部扰动的反应来估计。周期运动（如步行和跑步）附近的动力学和控制律将使用因式庞加莱图（经典庞加莱图的简单推广）来近似。因式庞加莱图是利用最大似然估计等统计技术从实验数据中推断出来的。这种推断的具体结果将允许预测人体如何在外部扰动的存在下恢复稳定运动，特别是，它将允许估计肌肉力量和身体运动是如何调节的，以恢复到稳定状态的循环步态。这些推断的控制规律将使用三维数学两足动物模型进行验证，以证明运动可变性的定量预测和跌倒可能性的准确估计。

Under normal circumstances, healthy adult humans can walk and run with a vanishingly small likelihood of falling. This level of assurance far exceeds the performance of state-of-the-art legged robots and robotic prosthetic devices. The goal of this project is to reverse engineer the dynamics and control of human locomotion using data from human subject experiments -- that is, to understand the control laws underlying human locomotion through observation of human subjects both walking naturally and responding to changes of terrain, sudden shoves, and other disturbances -- and to apply this understanding to more natural prosthetic devices and more capable legged robots. The results from the project will also give insight into movement disorders and other balance problems in the elderly and other at-risk populations, as well as allow design of more effective balance-improving devices. In this project, the physiological control laws governing human motion will be estimated, based on experiments from both natural unperturbed walking and running, and responses to carefully chosen external perturbations during locomotion. The dynamics and the control laws near periodic motions such as walking and running will be approximated using a factorized Poincare map -- a simple generalization of the classical Poincare map. Factorized Poincare maps are inferred from experimental data using statistical techniques such as maximum likelihood estimation. The specific results this inference will allow the prediction of how the human body will return to steady locomotion in the presence of a external perturbation, in particular, it will allow the estimation of how muscle forces and body movements are modulated to recover to steady state cyclic gait. These inferred control laws will be validated using three-dimensional mathematical biped models to demonstrate quantitative prediction of movement variability and accurate estimation of the likelihood of falls.