CAREER: Leveraging Toes for Energy-Efficient and Agile Humanoid Robot Locomotion

This Faculty Early Career Development Program (CAREER) award supports research to significantly enhance the locomotion capabilities of humanoid robots through the intelligent control of a novel articulated toe mechanism. Specifically, the project will show how a lightweight, powerful, and rugged mechanical toe can coordinate with existing actuators to dramatically improve walking efficiency, greatly increase running speed, and adapt a broad range of human-like movements to humanoid robots in challenging terrain and dynamic and unstructured environments. These results will contribute to national prosperity and public safety, by helping to realize the potential of humanoid robots to perform physically demanding or hazardous tasks for applications such as disaster response, infrastructure inspection, and other field operations. In addition, the project includes educational and outreach activities designed to attract new participants to robotics, including a year-long research program for high school students, a hands-on undergraduate course on robot control and learning, and summer workshops for K–12 learners that emphasize experiential engagement with engineering and technology. The project investigates efficient and versatile locomotion achieved through optimal utilization and coordination of articulated toes with other powered joints, such as knees, ankles, and hips. The following three critical challenges will be overcome using an imitation learning-based approach: First, discrepancies between human and robot dynamics are addressed through a bi-level trajectory optimization method that separates kinematic and dynamic planning, enabling the generation of diverse reference motions with automatic contact sequence selection. Second, a learning-based control framework is developed to achieve both optimality and versatility by incorporating optimality conditions into the training process, guiding the controller to effectively utilize toe actuation. Third, to bridge the gap between simulation and real-world deployment, a transfer strategy based on impedance matching is introduced, using frequency-domain analysis to capture system dynamics across a wide range of motion speeds. Together, these contributions are expected to advance the fundamental understanding of humanoid locomotion and provide generalizable methods for integrating mechanical design, optimization, and learning in robotic systems. This project is supported by the cross-directorate Foundational Research in Robotics program, jointly managed and funded by the Directorates for Engineering (ENG) and Computer and Information Science and Engineering (CISE). 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. NSF Award ID: 2541914 | Program: 01002627DB NSF RESEARCH & RELATED ACTIVIT | Principal Investigator: Donghyun Kim | Institution: University of Massachusetts Amherst, AMHERST, MA | Award Amount: $750,000 View on NSF Award Search: https://www.nsf.gov/awardsearch/show-award/?AWD_ID=2541914 View on Research.gov: https://www.research.gov/awardapi-service/v1/awards/2541914.html