China Reaches the Top in Robotics with Its Developed Maxwell AI Model
Developed by the Chinese Academy of Sciences, the artificial intelligence model named Maxwell ranked first globally in robotic physical tasks.
The new embodied artificial intelligence model named Maxwell, developed by the Artificial Intelligence Institute for Industries under the Chinese Academy of Sciences, achieved the highest score in the Meta-World test evaluating robotic physical tasks and rose to the first place.
Record Score in Meta-World Test
In the Meta-World simulation test established by Stanford University, UC Berkeley, and other institutions, robots' ability to perform fifty daily physical tasks is evaluated.
The Maxwell model achieved 91.9 points in this test, reaching the highest score ever recorded in the simulation test and placing first globally.
Other Models in Global Rankings
The FabriVLA model developed by Shenzhen-based Youibot ranked second with 90 points, while SUREFlow from Kyungpook National University in South Korea came in third with 88.3 points.
Wide Task Capacity and Success Rate
Maxwell was able to perform over two hundred tasks—including moving objects, adjusting positions, and operating switches and containers—without requiring additional fine-tuning.
Additionally, the model managed to achieve a ninety-nine point one percent success rate in the LIBERO simulation environment.
Technical Specifications and Industry Developments
Featuring one billion parameters, Maxwell supports single-card inference for deployment on edge devices.
Meanwhile, other Chinese organizations such as DeepCybo's PhysBrain model and Unitree Robotics have also made significant progress in physical foundation models and intelligent robot models.