I built a system to apply lateral balance perturbations to a human subject walking on an instrumented treadmill. The system centers around a microcontroller running step detection from treadmill force signals and PID tension control on two opposing motor-driven cable systems attached to a custom harness worn by the subject. A control computer loads the microcontroller with a profile of perturbations desired (force level, side, time during step) and receives streaming sensor data for logging. Without perturbations, the cables are held in light tension and tracks the lateral movement of the subject with minimal disturbance. Perturbations are introduced by increasing the net tension force level towards the side desired. The system was also adapted for a collaborative project to study stability during sit-to-stand movements.

Our understanding of human walking dynamics can be significantly informed by simple walking models derived from inverted pendulum-like dynamics with ground collision energetics. I implemented a 3D walking model with multiple simplified actuators to represent possible balance control mechanisms and demonstrated the varying controllability of balance stability using classical control theory and optimization approaches.

I conducted a number of balance perturbation experiments on human subjects to understand how people might balance besides controlling their step placement. I was responsible for the entire experimental process, from experimental design, piloting, ethics review, apparatus design and construction (see perturber above), and data collection, cleaning, and analysis.

I created a number of software tools for time-series motion and force data cleaning, event detection, motion and force data integration, and inverse dynamics analyses. Methods included automation with thresholding, supervised learning classification, and matrix profile and dynamic time warping combined with manual interventions.

Underactuated bioinspired robots using smart composite microstructures (SCM) could create a running gait reminiscent of many-legged robots like RHex, but had difficulty matching the energetic efficiency and multi-substrate performance capabilities of the larger machine. I created a polymer-casting process for making flexibly C-shaped legs whose stiffness could be modulated with shape-memory alloy (SMA) actuators to take advantage of passive spring dynamics and improve performance.