Stas Tiomikin, has been awarded a NSF grant to support research aimed at advancing robotic intelligence.
Stas Tiomkin, assistant professor in the Department of Computer Science at Texas Tech University’s Edward E. Whitacre Jr. College of Engineering, has been awarded a $105,524 grant from the National Science Foundation (NSF) to support cutting-edge research aimed at advancing robotic intelligence through the development of self-recovering systems.
The project seeks to break new ground in understanding viability in robotics, the capacity of a system to autonomously maintain or regain functionality following a critical failure. From underwater drones stuck in unrecoverable positions to robotic limbs immobilized in cluttered environments, the concept of viability addresses how artificial agents can act independently in unpredictable or dangerous conditions.
“This multidisciplinary project pushes the boundaries of robotic intelligence by exploring the viability of robots,” Tiomkin said. “I’m thrilled to explore this cutting-edge field, where each discovery unlocks new possibilities for building autonomous robotic systems that are not only powerful, but also efficient and interpretable. Receiving this award strengthens my commitment to advancing robot learning and driving innovation in next-generation intelligent systems.”
Tiomkin’s research will focus on developing a framework and metric to measure an agent’s viability using data derived from its own sensors and motors. The framework will first identify fundamental control system properties to define and calculate the viability metric efficiently. It will then be tested in two real-world applications: the self-recovery of walking robots and the self-maintenance of vehicle braking systems.
A core innovation in the project involves extending the classical mathematical notion of Lyapunov exponents into what Tiomkin terms Agent-Induced Lyapunov Exponents (AILE). This approach prioritizes states with higher adaptability and greater potential for successful self-control. Unlike traditional machine learning techniques that require externally defined goals or rewards, AILE would allow robots to learn and adapt autonomously across a wide range of scenarios.
The broader implications of the research span beyond robotics into the domains of machine learning, control theory, mechanics and information theory. By crossing disciplinary boundaries, the project aims to fuel the creation of next-generation intelligent systems capable of operating independently in high-risk or critical environments, including autonomous driving and disaster response.