DaxBrain-WM (Dualis)

DaxBrain-WM, marketed as Dualis, is the control model DAX Robotics announced as commercially deployed at WAIC in Shanghai in July 2026 and now ships on its Qiji rideable machines. Its distinguishing idea is a deliberate split into two halves, named after the yin-yang pair of Chinese cosmology. The yin half is a multimodal mixture-of-experts vision-action model handling the lowest control loop: it fuses vision, touch, force and proprioception and is stated to respond in under 50 ms. The yang half is a physics-driven planner working on explicit rules, responsible for breaking a task into steps, scheduling across layers, watching for anomalies and recovering from them, and coordinating several machines at once. DAX frames this as a rejection of pure end-to-end vision-language-action systems, which its CTO argues demand too much data, cost too much to train and generalise too weakly to deploy commercially; the layered alternative is claimed to shrink the search space for ordinary tasks by orders of magnitude. Everything runs on the robot, without a cloud connection, so gait planning and joint torque allocation happen locally in milliseconds. Readers should note what is missing: DAX publishes no parameter count, no training-data description, no benchmark results and no weights or API. This entry describes an architecture the company has explained, not a model anyone outside DAX can test.

▸Videos

No videos yet.

▸Reviews