Research Overview

Movement science is the study of how humans and animals represent, organize, modify, and execute actions. It seeks to answer questions such as: How do we control our bodies so nimbly? How do we learn new motor skills? And how do we then adapt these skills to vastly different circumstances? To study these phenomena we employ our engineering skills in a normative framework; we build computational models that quantify how the nervous system should optimally produce motor behaviors. These models rely on diverse fields ranging from optimal control theory, decision theory, Bayesian statistics, machine learning, neuroscience and muscle physiology, to prescribe how best to achieve the motor behavior under question. Then we compare our models with experimental evidence. This in turn provides insight into the nervous system's goals, and a conceptual framework for how the nervous system achieves those goals.

Ongoing Projects

Subjective Motor Costs 
How do the subjective costs of our motor system influence our movements, and how do these costs change with age? Together with our collaborators in Alaa Ahmed's group, we are examining these issues, building computational models to explain learning and control in terms of optimal policies.

Sensorimotor Decision-making
How do people make motor choices when confronted with uncertain information, and how do we represent these uncertainties? We examine these questions with our collaborators in Konrad Kording's group, using experiments and Bayesian models to quantify human behavior.

Deep Networks for Motor Control
How does the nervous system generate commands for movements, and how are new commands learned? We are exploring new advances in machine learning to build self-organized, low-dimensional representations of command trajectories.

Motor Generalization
How does the nervous system represent motor information? For example, are new behaviors represented parametrically, or like a look-up table? To address this question we use computational models and experiments to examine how newly learned motor behaviors generalize.

Control of Noisy Muscles
How should the nervous system, or an artificial controller (e.g. FES), command muscles when their output is inherently uncertain? Through optimal stochastic control and animal studies (with Matthew Tresch) we are exploring these issues. Our results offer a proof of concept for FES.

Motor Adaptation 
When we adapt, what are we learning? We should be learning about both the world and our body. Using the known phenomena of adaptation we explore whether or not the nervous system represents both our bodies and the world.

Previous Projects

Powered Prosthetics
How do you control a prosthetic limb to behave just like your own? In this work we designed an EMG-driven controller for the world's first powered ankle prosthetic.

Low-dimensional Control
Does the nervous system use low-dimensional representations of our body to simplify control? In this work we demonstrate how a balanced and truncated model produces control similar to the full-dimensional optimal solutions.