Jan Ole von Hartz

Papers

Total Citations

H-Index

About

Jan Ole von Hartz is a roboticist at the forefront of sample-efficient robot learning, with his work bridging the critical gap between perception and policy. His research centers on enabling robots to learn complex, long-horizon manipulation tasks from minimal human input, tackling the fundamental challenge of data scarcity in robotics. In his highly cited work, "The Treachery of Images" (10 citations), he introduced Bayesian Scene Keypoints, a method that extracts compact, scale-invariant representations from camera data, dramatically improving sample efficiency for deep policy learning in manipulation. Building on this, his "Art of Imitation" paper (7 citations) advanced Task-Parametrized Gaussian Mixture Models (TP-GMMs) to handle real-world challenges like varying object poses and velocities, enabling robots to learn from just a handful of demonstrations. Most recently, his work on "Whole-Body Teleoperation" (3 citations) proposes a zero-cost framework for collecting demonstration data from mobile manipulators, a critical bottleneck for training robotic foundation models. By addressing the "treachery" of visual ambiguity and the practical hurdles of data collection, von Hartz is paving the way for robots that can learn dexterous, real-world skills with unprecedented efficiency.