Nikolay Serbenyuk

Papers

Total Citations

H-Index

About

Nikolay Serbenyuk is a roboticist specializing in autonomous systems for heavy machinery, with a core focus on learning-based control for field robotics. His research bridges imitation learning and reinforcement learning to tackle complex, real-world manipulation tasks, particularly in unstructured environments. Serbenyuk’s major contribution lies in developing methods that enable autonomous robotic wheel loaders to master pile loading—a highly dynamic and sensor-rich task. His 2020 paper, “Learning a Pile Loading Controller from Demonstrations,” introduced a controller that learns from just a handful of human demonstrations, leveraging low-level sensor data (boom angle, bucket angle, hydrostatic pressure) and egocentric video to replicate expert behavior. This work, with 7 citations, laid the groundwork for data-efficient learning in heavy equipment. Building on this, his 2022 paper, “Visual Rewards From Observation for Sequential Tasks: Autonomous Pile Loading,” addressed a critical bottleneck in reinforcement learning: reward function design. By deriving visual rewards directly from observation, Serbenyuk enabled autonomous learning without abundant datasets or extensive training time, a breakthrough for field robotics where environmental variation is high. With 5 citations, this work highlights his impact on making RL practical for real-world applications. Serbenyuk’s research is pivotal for advancing automation in construction, mining, and agriculture.