We are thrilled to announce that the paper 'The Treachery of Images: Bayesian Scene Keypoints for Deep Policy Learning in Robotic Manipulation' authored by Jan Ole von Hartz, Eugenio Chisari, Tim Welschehold, Wolfram Burgard, Joschka Boedecker, and Abhinav Valada, has been honored with a Best Paper Award Honorable Mention from the IEEE Robotics and Automation Letters (RA-L). The paper was selected as one of the top 10 papers in 2023 from over 1,200 published works, marking a significant achievement for the team and recognizing the impact of the research. The work was partly funded by the center BrainLinks-BrainTools.
The award was presented at this year’s IEEE International Conference on Robotics and Automation (ICRA) held in Yokohama, Japan. ICRA is the flagship conference of the IEEE Robotics and Automation Society (RAS), attracting more than 7,000 attendees from around the world.
The paper addresses two crucial challenges in learning robot policies from camera observations: sample efficiency and generalization. The team developed BASK—a Bayesian approach for tracking scene keypoints over time. Learning policies from these scene keypoints is significantly more efficient than learning directly on images. And the keypoints generalize across objects and environments, while being robust to disturbances.
BASK marks an important step from getting robot learning out of the lab, and into diverse environments, such as homes and factories.
We are grateful to the IEEE Robotics and Automation Society for this honor and to all our collaborators and supporters who have been a part of this journey. We look forward to further advancing our research and exploring new frontiers in robotic perception and manipulation.
Link to the original work: http://bask.cs.uni-freiburg.de/