UC Berkeley Researchers Win Top Honors at World’s Leading Robotics Conference
A team of researchers from UC Berkeley has received the Best Paper Award on Robot Learning at the 2025 IEEE International Conference on Robotics and Automation (ICRA), considered the world’s premier conference in robotics and automation. The event was held in Atlanta 19-23 May.
The award went to UC Berkeley IEOR Professor Ken Goldberg, his IEOR PhD student Lawrence Yunliang Chen, and their collaborators in the lab and at GoogleDeepmind for their work on Robo-DM, a new open-source toolkit for robot data management, designed to make large-scale robot learning more practical and efficient.
As artificial intelligence researchers increasingly turn to transformer-based models to teach robots new skills, the need for massive datasets of robot demonstrations has grown. These datasets, which often include high-bandwidth video, text, and numerical streams from multiple cameras, present a major technical bottleneck when it comes to generation, storage, distribution, and training.
Robo-DM addresses this challenge by providing a cloud-based data management system that compresses, stores, and retrieves robot trajectory data with high efficiency. The toolkit uses the Extensible Binary Meta Language (EBML) format to significantly reduce dataset size—up to 70 times smaller with lossy compression—while preserving performance during downstream tasks such as pick-and-place operations. In benchmark comparisons, Robo-DM outperforms other frameworks, offering faster decoding and more efficient data loading during training.
The research shows that even with aggressive compression, models trained using Robo-DM maintain task accuracy. The authors believe the toolkit could accelerate progress in robotics by making it easier to train flexible, generalizable systems on large and diverse datasets.
ICRA 2025 received over 4,000 submissions. The recognition of Robo-DM underscores the increasing importance of scalable infrastructure for training the next generation of intelligent robotic systems.
This award also coincides with a broader surge in public interest around robotics and artificial intelligence, reflecting the field’s expanding relevance beyond academia and industry. Ambi Robotics, the company professor Ken Goldberg c-ofounded in 2018, also received an award at the same conference for Most Innovative Product. Ken has recently brought these conversations to a broader audience through high-profile podcast appearances, including Armchair Expert with Dax Shepard—which reaches over 20 million listeners monthly—and the Freakonomics podcast. In these episodes, he discusses the future of robotics, the complexities of robotic grasping, and how art and technology continue to inform each other in unexpected ways. By bridging technical innovation and public discourse, Goldberg is helping demystify robotics at a time when societal understanding is critical to responsible adoption.
Learn more about ICRA 2025