Rutgers Department of Computer Science is pleased to share that Professors He Zhu, Kostas Bekris, and Abdeslam Boularias are the recipients of an NSF award for their project "FMitF: Track I: Abstraction Refinement-guided Program Synthesis for Verifiable Robot Learning".
This project advances science at the intersection of robotics, artificial intelligence, and formal verification to enable reliable and transparent robot behavior in real-world settings. As robots increasingly assist with complex tasks, from warehouse logistics to supporting independent living, ensuring their safe, trustworthy operation is essential. However, state-of-the-art robot learning methods, such as deep reinforcement learning, rely heavily on opaque neural network controllers that are difficult to interpret, verify, and generalize, limiting their use in safety-critical domains. This research addresses these challenges by developing a new class of interpretable control programs, written in domain-specific languages with automatically-inferred domain knowledge. These programs enable robots to reason over long-term goals, adapt to novel environments, and be certified as safe before deployment. Through this approach, the project aims to improve both the efficiency and reliability of real-world robotic systems. *
This is a three-year award with a funding amount of $900K.
.jpeg)