A Scalable Platform for Robot Learning and Physical Skill Data Collection

Publication
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

This is a system paper describing the concepts, architecture, and implementation of the Collective Learning System, which in this paper we termed as Parallel and Distributed Robot AI (PD.RAI).

The collective learning project has been conceptualized by Prof. Sami Haddadin long time ago. After the founding of MIRMI institute (formerly called MSRM), our previous colleagues, mainly Lars Johannsmeier started the early work using about 10 robots.
In KI.Fabrik project we had the opportunity to make a bigger version. The project was delayed by the Pandamic, and since 2022 we started to rent a basement area of the beautiful Deutsches Museum and converted it into a lab space. In 2023 I led my team, Samuel Schneider and Yansong Wu, with the help of many colleagues including Lingyun Chen, Mehmet C. Yildirim, Mohamadreza, Zheng Shen, to establish this robot cluster, originally 50 dualarm robots, made of 100 Franka Emika Panda arms. The first public showcase of this collective learning robot cluster was in the 2023 AI.BAY conference in Munich. Since then it has attracted a lot of attention from the public and media, and been visited by many scientists, politicians, and even school kids.

This video was showcased during the ICRA 2024 conference when Prof. Sami Haddadin was giving the plenary talk The Great Robot Accelerator: Collective Learning of Optimal Embodied AI. Note that TUM MIRMI owns the copyright of the video, no other individual or company has the right to use it without permission, except the contributors I mentioned above.