Roman Zashchitin
Papers
2
Total Citations
10
H-Index
2
About
Roman Zashchitin is forging a path at the intersection of reinforcement learning and robotics, with a sharp focus on creating intelligent agents that are both autonomous and provably stable. His core research addresses a fundamental challenge: how to enable mobile robots to navigate and learn in unknown environments without relying on pre-built maps. His 2023 work on "Predictive reinforcement learning" introduced a map-less navigation method that has already garnered 8 citations, signaling its practical value for real-world robotic deployment. Zashchitin’s most notable contribution is the development of the **Critic As Lyapunov Function (CALF)** agent, presented in 2024. This model-free reinforcement learning algorithm is a breakthrough in safety-critical control. By ingeniously using the critic network itself as a Lyapunov function, CALF guarantees that the environment—or dynamical system—is stabilized *during every learning episode*, not just after convergence. This online stability assurance is a significant leap from traditional RL, which often allows destabilizing exploratory actions. With 2 citations in its first year, this work is poised to influence the design of reliable autonomous systems. Zashchitin’s research is essential reading for anyone interested in bridging the gap between theoretical control theory and practical, learning-based robotics.
Research Focus
Key Achievements
Top Papers
- 1
- 2