Alexander Zhurkevich

Papers

Total Citations

H-Index

About

Alexander Zhurkevich is a leading researcher in robotic manipulation, with a primary focus on bridging the gap between simulated training environments and real-world dexterous control. His most impactful contribution is the development of **DeXtreme**, a framework that demonstrates how deep reinforcement learning can enable agile, multi-fingered in-hand manipulation—such as spinning a pen or reorienting objects—trained entirely in simulation and then transferred to physical robotic hands. This work, which has accumulated over 90 citations across its iterations, addresses the critical challenge of the sim-to-real gap, where policies learned in virtual environments often fail in practice due to modeling inaccuracies. Zhurkevich’s approach combines robust domain randomization with carefully designed reward structures, achieving unprecedented levels of dexterity and speed in real-world tasks. His research is notable for its practical impact on robotics, offering a scalable pathway for developing complex manipulation skills without extensive real-world data collection. By advancing the reliability of sim-to-real transfer, Zhurkevich is helping to unlock the potential of dexterous robots for applications in manufacturing, assistive technology, and beyond.