Gleb Shevchuk

Papers

Total Citations

H-Index

About

Gleb Shevchuk is a robotics and machine learning researcher whose work centers on two intertwined challenges in autonomous systems: enabling robots to learn without hand-crafted reward functions, and efficiently capturing human intent to guide robot behavior. His research sits at the intersection of reinforcement learning, reward learning, and visuomotor control — areas critical to making robots genuinely useful in unstructured, real-world environments. One of Shevchuk's most recognized contributions is his work on Distributional Planning Networks, which tackles the costly problem of manual reward engineering by enabling unsupervised visuomotor control — allowing robots to learn goal-directed behavior directly from visual input without task-specific supervision. This work has garnered over 40 citations across related publications. Equally impactful is his research on integrating human demonstrations and preferences to learn reward functions, published in both 2019 and an expanded 2021 version accumulating nearly 40 combined citations. These works offer principled frameworks for optimally combining passive and active human feedback, bridging inverse reinforcement learning with preference-based approaches. Shevchuk's contributions speak to a broader mission: reducing the human engineering burden in robotics while keeping humans meaningfully in the loop — a balance essential for the next generation of intelligent, adaptive machines.