New Chinese embodied AI model Maxwell tops global physical-task benchmark
The Maxwell embodied artificial intelligence model has taken first place in the global Meta-World benchmark for robotic physical-task performance, scoring a record 91.9 points.
Gurban Gurbanzadeh1 min read
A new Chinese embodied artificial intelligence model, Maxwell, developed by the Institute of Artificial Intelligence for Industry at the Chinese Academy of Sciences, has taken first place in the global Meta-World benchmark, which evaluates how well robots perform physical tasks.
Maxwell scored 91.9 points — the highest result in the history of this simulation benchmark.
Meta-World was created by researchers from Stanford University, the University of California, Berkeley and other institutions. It assesses how well robots handle 50 everyday physical tasks — from grasping and moving objects to opening doors and using boxes.
Leading teams from around the world submit entries to the ranking, including Physical Intelligence, Google DeepMind, Alibaba, Meituan, Carnegie Mellon University, MIT, the University of Cambridge and the Toyota Research Institute.
Who follows the leader
Among the latest submissions filed in July, second place went to the FabriVLA model, developed by the Shenzhen-based company Youibot, with a result of 90 points. Third was SUREFlow from researchers at Konkuk University in South Korea, with 88.3 points.
Maxwell thus not only topped the physical-task ranking but also improved the best result by several points over its closest competitors from China and South Korea.
Source: scmp.com
