We create mathematical models that capture essential principles of legged locomotion with a focus on dynamic gait stability, balance recovery, and agility. These simplified models reveal core mechanical principles that enable stable and efficient locomotion in any legged system, whether human or humanoid.

Simple models provide clear insight but often lack the complexity needed for real-world control. We establish principled frameworks for systematically decomposing complex control systems. These decomposition frameworks enable automated discovery of hierarchical controllers and offer a physics-informed alternative to pure machine learning approaches. The methodology extends beyond locomotion to other robotics domains including manipulation and flight.

We investigate how the human neuromuscular system controls and learns locomotion. Our work bridges biomechanics and neuroscience. We seek to understand how (bio)mechanical principles of legged locomotion connect to neural control and how this control adapts through learning. Knowledge about these fundamental neural mechanisms can help explain the effects of aging and damage to the motor system on human gait, and inform rehabilitation strategies after spinal cord injury and stroke. This knowledge also produces human-like control strategies for powered leg prostheses, exoskeletons, and legged robots.

We explore novel control strategies for powered lower-limb assistive robots that can improve human mobility and quality of life. Our research combines human-like neuromuscular control strategies with machine learning approaches to create artificial lower limbs that continuously reason about and adapt to their human user and the environment, advancing technologies for safer, more agile, and natural locomotion in real-world conditions.