Raz Yerushalmi

Papers

Total Citations

H-Index

About

Raz Yerushalmi is a rising researcher at the intersection of robotics, artificial intelligence, and formal verification, with a primary focus on ensuring the safety and reliability of learning-based autonomous systems. His work tackles a critical challenge: while deep reinforcement learning (DRL) enables robots to master complex behaviors, the resulting policies are notoriously fragile and prone to bugs. Yerushalmi’s major contribution lies in developing verification methods specifically tailored for DRL-driven robotic navigation, demonstrating how formal techniques can expose failures in learned controllers before deployment. His most-cited paper (2023, 18 citations) is a pioneering step in this direction, bridging the gap between DNN verification research and real-world robotic systems. In parallel, his work on constrained reinforcement learning (2022, 4 citations) introduces scenario-based programming to embed safety constraints directly into the learning process, ensuring that robots optimize performance without compromising safety. By combining rigorous verification with practical robotics, Yerushalmi is helping to pave the way for trustworthy autonomous systems in safety-critical applications—a mission that resonates deeply as robots increasingly operate alongside humans.