Anne Martin


Due to experimental difficulties, almost no scientific evidence to date definitively indicates that one lower-limb prosthesis performs better than another. A model of walking that is simple enough to allow systematic exploration of prosthesis design variables, yet detailed enough to accurately capture step dynamics, could help fill this gap. Further, such a model could be used to design optimal prostheses. Unfortunately, an appropriate model does not yet exist, even when the problem is simplified to a healthy subject. To create an appropriate model, a robotics modeling and control technique called Hybrid Zero Dynamics (HZD) can be extended to create a planar model of human walking.

The first improvement was to add feet to the model. During normal walking, human feet function as circular arcs. My work focused on modeling the step-to-step transition and extending the HZD control law to capture the changes due the rolling foot contact. Replacing the point feet common to robots controlled with HZD with curved feet dramatically reduced the energy required for walking in both simulation and experiment on the biped ERNIE.

The second feature that I added was toe-off. In human walking, the ankle generates a significant amount of power, particularly at the end of stance and this ankle power is called toe-off. It wasn't clear initially if toe-off would occur passively due to the curved foot or if an ankle joint needed to be included in the model. I created two models - a six-link model with ankles and a four-link model without ankles, and tried to match the models' motion to human walking as closely as possible. Note that because I am modeling the function of the human foot and ankle, and not the form, the ankle joints are located on the diameter of the circle, which makes the feet look very large. The results of the study were clear - ankle joints were needed to allow the model to match both the kinematics and energetics of human walking.

Although there are many possible ways for a human to coordinate his/her joints that results in walking, healthy humans all choose approximately the same motion. This indicates that walking is optimal in some sense, which means that if the correct objective function is found, the six-link model can predict how a human will walk. A torque-squared-based objective function works very well, and can predict both the kinematics and energetics of walking at speeds ranging from very slow to almost running.

If the six-link model is modified so that one side has a passive or fixed ankle, this model can be used to predict the way amputees will walk. This prediction can then be used in optimizing the design of a prosthesis and in determining how to correctly align the prosthesis to the residual limb.

Journal Papers

Conference Publications