Xupeng Zhu

Papers

Total Citations

H-Index

About

Xupeng Zhu is a robotics researcher whose work lies at the intersection of manipulation, learning, and geometric deep learning. His key research areas include robot grasp learning, contact force estimation, and the development of equivariant models that exploit spatial symmetries to improve data efficiency. Zhu’s major contributions include pioneering the use of equivariant neural networks for robotic manipulation, as demonstrated in his work on equivariant Q-learning in spatial action spaces (4 citations) and robot grasp learning (9 citations). He also introduced "Forces for Free" (7 citations), a vision-based method for contact force estimation that eliminates the need for fragile, expensive force/torque sensors. Additionally, Zhu co-developed BulletArm, an open-source robotic manipulation benchmark and learning framework (5 citations), providing a standardized platform for the research community. His work addresses fundamental challenges in real-world robotics—such as hardware noise, data scarcity, and sensor fragility—by leveraging symmetry and compliant hardware. With a growing citation impact and a focus on practical, sensor-free solutions, Zhu is shaping the future of adaptive, data-efficient robotic systems.