Carl Doersch

Papers

Total Citations

H-Index

About

Carl Doersch is a leading researcher at the intersection of computer vision and robotics, with a primary focus on enabling machines to learn from visual data with minimal human intervention. His work centers on self-supervised learning, visual correspondence, and few-shot imitation—pushing the boundaries of how robots can acquire new skills efficiently. Doersch is perhaps best known for his foundational contributions to unsupervised visual representation learning, including the highly influential "Context Prediction" paper, which has amassed over 1,300 citations and helped pioneer the use of spatial context as a supervisory signal. More recently, his work on "RoboTAP" (2024) introduces a novel method for tracking arbitrary points in video to enable few-shot visual imitation, allowing robots to learn new behaviors from just a single demonstration without task-specific engineering. This approach represents a significant leap toward practical, general-purpose robotics. With a career spanning top venues like CVPR, NeurIPS, and ICLR, Doersch’s research consistently bridges the gap between theoretical advances in self-supervised learning and real-world robotic applications, making him a key figure in the drive toward adaptable, data-efficient autonomous systems.