Idan Shenfeld

Papers

Total Citations

H-Index

About

Idan Shenfeld is a leading researcher in robot learning, specializing in imitation learning and reinforcement learning for precise, long-horizon manipulation. His work directly tackles the central challenge of enabling robots to perform high-precision assembly tasks without requiring massive demonstration datasets. In his highly cited paper "JUICER: Data-Efficient Imitation Learning for Robotic Assembly" (2024, 21 citations), Shenfeld introduced a novel pipeline that dramatically improves imitation learning performance for complex visuomotor policies, proving that robots can learn from far fewer examples than previously thought. Building on this, his follow-up work "From Imitation to Refinement - Residual RL for Precise Assembly" (2025, 7 citations) identifies a critical bottleneck: behavior cloning performance saturates with increasing data due to fundamental limitations in precision. Shenfeld's key contribution is the development of residual reinforcement learning frameworks that refine initial imitation policies, overcoming this saturation to achieve reliable, high-accuracy assembly. His research is shaping the future of industrial robotics by making robot teaching more practical and scalable, directly impacting how robots learn to perform delicate, real-world tasks.