EECS researchers win Best Robot Learning Paper Award at IEEE ICRA 2025

Professors Ken Goldberg, John Kubiatowicz, and their team of students and collaborators: Kaiyuan Chen, Letian Fu, David Huang, Yanxiang Zhang, Lawrence Yunliang Chen, Huang Huang, Kush Hari, Ashwin Balakrishna, Ted Xiao, and Pannag R Sanketi, have received the Best Robot Learning Paper Award at the IEEE International Conference on Robotics and Automation (ICRA) 2025 in Atlanta, GA.

The group was cited “for valuable publicly available data management tools that facilitate efficient storage and access of large-scale multimodal data, which is critical for robot learning.”

Their paper, “Robo-DM: Efficient Robot Big Data Management,” addresses the growing challenges of managing large, multimodal datasets in robot learning, including video, language, and sensor data, by introducing a highly compressed, unified file format using Extensible Binary Meta Language (EBML). The toolkit enables faster data loading, plug-and-play integration with popular training frameworks, and significant reductions in storage and transfer costs (up to 75x compression). Experimental results show Robo-DM supports training with minimal impact on performance, even under aggressive compression.

This work reflects Berkeley’s leadership at the intersection of robotics, data systems, and AI, and the team’s contribution promises to advance how large-scale robotic learning datasets are curated, shared, and deployed in both academia and industry.