OASIS WAM, Built by Synkrotron and USTC, Tops RoboLab Global Simulation Benchmark
In September 2026, RoboLab, a global high-fidelity simulation benchmark platform, released its latest rankings. OASIS WAM, developed by Synkrotron in collaboration with the University of Science and Technology of China, took first place overall in a comprehensive evaluation covering 120 vision-language-manipulation tasks. The result placed it ahead of leading embodied AI teams including NVIDIA, Google, and Physical Intelligence, making it the only Chinese team on the list.
RoboLab is built on NVIDIA Isaac and NVIDIA Omniverse simulation technologies, providing high-fidelity visual environments and physics-based modeling for large-scale training and testing of robot policies. According to the platform, OASIS WAM's ranking was confirmed through actual tests via open API calls, not self-reported offline results. RoboLab uses dynamically generated scenes, variable language instructions, and controlled environmental perturbations, with no fixed training set.
OASIS WAM is described as a world action model for general-purpose robots. It uses a layered architecture in which a high-level vision-language model (VLM) interprets tasks and selects the next manipulation subgoal, while WAM generates continuous actions and predicts future visual states during execution. A lightweight Speculator model can also predict the VLM's next command in advance, allowing the robot to start generating subsequent actions before the VLM finishes reasoning, thereby reducing waiting time.
Synkrotron, a Chinese startup incubated by Turing Award laureate Andrew Yao, said it plans to build an embodied AI infrastructure around OASIS WAM, a hybrid data platform, and Dora OS, aiming to support applications such as medical research, industrial manufacturing, and inspection, while remaining compatible with multiple robot forms including manipulators, humanoid robots, and quadruped robots.
