EPSCoR Research Fellows: @NSF: Integrating Physical Principles and Learning for Safe and Reliable Locomotion of Legged Robots

This Research Infrastructure Improvement (RII) EPSCoR Research Fellows project provides a fellowship to an Assistant Professor and training for a graduate student at the University of New Mexico. This work is conducted in collaboration with Aaron D. Ames at the California Institute of Technology. Through the fellowship, the PI will develop a new framework to help legged robots move safely and efficiently in hazardous environments characterized by rough and changing terrain. The framework will integrate machine learning with physical principles and embedded safety mechanisms, enabling robots to select appropriate movement patterns, such as walking, trotting, or running, while detecting and correcting unsafe or unstable actions before they are executed. By making learned robot behaviors more adaptable, interpretable, and reliable, the project will advance robotic capabilities for disaster response, infrastructure inspection, environmental monitoring, and search-and-rescue operations. It will also strengthen robotics education at the University of New Mexico through new courses, student competitions, K–12 outreach, academic events, and industry partnerships. Legged robots provide critical mobility in hazardous environments that may be inaccessible to human workers. A central challenge in advancing legged-robot locomotion is developing controllers that combine high performance and reliability with systematic gait adaptation across changing terrain. This project will address this challenge through an integrated framework that combines physical modeling, advanced control theory, and machine learning. The research comprises three synergistic thrusts: a differentiable optimization layer that projects neural-network-generated actions onto sets satisfying formal safety and stability requirements; an interpretable gait-selection module that adapts locomotion modes to terrain conditions and task demands to improve stability, mobility, and energy efficiency; and GPU-accelerated parallel solvers that enable high-frequency, real-time control. The project will also strengthen robotics and machine-learning infrastructure at the University of New Mexico through faculty development, student training, new robot-learning curricula, and expanded simulation and experimental capabilities. These efforts will further support student competitions, K–12 outreach, academic events, and industry partnerships. This project is supported by the EPSCoR Research Infrastructure Improvement Program: EPSCoR Research Fellows (ERF). The ERF program supports early- and mid-career investigators in eligible jurisdictions to develop collaborations at the nation’s private, government or academic research institutions. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.