At the 2025 Beijing Intelligence Conference held today, Beijing Zhiyuan Artificial Intelligence Research Institute officially released the embodied intelligent operating system RoboOS 2.0 and the upgraded large model RoboBrain 2.0, both fully open-source, promoting an open and collaborative embodied intelligence ecosystem.
RoboOS 2.0 is the first robot operating system to support the MCP (Multi-Entity Collaborative Platform) mechanism. It features serverless entity deployment capabilities and a no-adaptation cerebellum skill registration mechanism, significantly reducing development barriers. In terms of system performance, the response latency has been compressed to below 3 milliseconds, and the end-to-end cloud communication efficiency has increased 27 times. Additionally, it introduces a spatiotemporal scene graph sharing and task closed-loop feedback mechanism, greatly enhancing multi-robot collaboration capabilities and task stability.
The accompanying upgraded RoboBrain 2.0 is an open-source embodied brain large model with capabilities for dynamic modeling of multi-entity environments and cross-entity collaborative task planning. Test data shows that its task planning accuracy has improved by 74% compared to the previous generation, significantly surpassing existing mainstream models in spatial reasoning and intelligent scheduling.
Currently, both RoboOS 2.0 and RoboBrain 2.0 have been fully open-sourced, including framework code, model weights, datasets, and evaluation benchmarks. Zhiyuan stated that it has collaborated with more than 20 embodied intelligence companies globally to jointly build an open, thriving, and collaborative intelligent robotics ecosystem.