Westlake Robot has significantly upgraded its general-purpose robotic brain, WR1, by implementing the world's pioneering dual pre-training system architecture that coordinates the functions of the brain and cerebellum. This innovative architecture seamlessly combines a world model with a visual-language-action model, and is regulated by the world's first general-purpose action model, GAE, to enable comprehensive full-body control. WR1 has matched or even outperformed international benchmarks in several key capabilities, successfully accomplishing demonstration tasks such as agile clothing manipulation and coordinated household chores across diverse environments. This marks a breakthrough in overcoming the original technical limitations of humanoid robots. The underlying technical prowess is attributed to three core foundational capabilities: the ability to generalize handling of flexible objects, execute cross-scenario long-term tasks, and integrate fluid motion with manipulation. These capabilities arise from the system architecture's integration across multiple dimensions, including body parts, time dimensions, intention and execution, as well as end-to-end processes. As a commercialization project leveraging relevant achievements from Westlake University, Westlake Robot adheres to a trinity full-stack self-developed approach that encompasses a general-purpose brain, a humanoid full-body cerebellum, and a self-developed humanoid body. It aims to bridge the gap between task understanding and full-body motion control, thereby facilitating the application and implementation of humanoid robots in various scenarios.
