The field of embodied intelligence has made a milestone breakthrough. Gaode officially announced the full open-source release of the world's first robot embodied operation base model based on a unified architecture, ABot-M0. The model is designed to achieve "a general brain that adapts to various robot forms," aiming to break down barriers between heterogeneous hardware and accelerate the transition of embodied intelligence from laboratories to industrial and home environments.
Core Technologies and Performance
ABot-M0 has demonstrated outstanding performance in multiple industry authority benchmark tests. Data shows that the model achieved an impressive task success rate of 80.5% on the Libero-Plus benchmark, which is nearly 30% higher than the previous industry standard solution Pi0. In addition, it set new SOTA (industry-leading) records in tests such as Libero and RoboCasa.
Full Open-Source in Three Dimensions
To address long-standing issues in the field of embodied intelligence, such as "data silos" and "deployment difficulties," Gaode has open-sourced three dimensions: underlying data, core algorithms, and pre-training models:
Data Level: Open-sourced the largest general-purpose robot dataset currently available, UniACT. This dataset integrates over 6 million real operation trajectories and provides a complete processing pipeline from heterogeneous data to standardized training data.
Algorithm Level: Also released the model architecture and training framework. Key highlights include Gaode's innovative Action Manifold Learning (AML) algorithm and two-stream perception architecture, endowing robots with excellent spatial understanding and action execution capabilities.
Model Level: Provides end-to-end pre-training models and a complete toolchain. Developers can use them "out of the box" without building the framework from scratch, significantly lowering the barrier for adapting to industrial collaboration or home service robots.
Industry Impact
Gaode's ABot-M0 technical leader stated that true general embodied intelligence requires joint refinement by global developers. The open-source release of ABot-M0 is not only about sharing technology but also aims to build a bridge connecting academic research and industrial application, enabling every robot of different forms to have a smart, reliable, and general "brain."
