Gasgoo Munich-On June 15, AI² Robotics Founder and CEO Guo Yandong took the stage at the 2026 Beijing Zhiyuan Conference. He unveiled NeuroVLA, the world’s first brain-inspired embodied intelligence system. Guo announced that the system is now open source on the AlphaBrain Platform.
Developed jointly by AI² Robotics and the Hong Kong University of Science and Technology (Guangzhou), the NeuroVLA architecture mimics the human brain’s three-layer system: cortex, cerebellum, and spinal cord. The cortex handles deep reasoning and planning, the cerebellum ensures dexterity and stability, while the spinal cord enables instinctive, rapid reactions.
NeuroVLA is now available on AI² Robotics’s AlphaBrain Platform, a one-stop open-source community for embodied intelligence, accessible to the entire industry.
Image Credit: AI² Robotics
Gasgoo has observed that amid industry debate over whether "VLA is obsolete," Guo has reframed the question. He argues that any intelligent system operating in the physical world requires three core capabilities: perception, reasoning, and control. These elements — vision, language, and action — are eternal, he says; only the way they are organized evolves.
Consequently, the paradigm debate isn’t about replacement but about organizational evolution. New technologies like world models and brain-inspired architectures don’t overturn VLA; they enhance and complete it. "VLA won’t vanish," Dr. Guo emphasized at the event. "It will be continuously empowered, becoming ever smarter. It remains the strongest pathway to intelligence in the physical world."
Guided by this philosophy, AI² Robotics has mapped out a three-stage roadmap for VLA. It begins with the end-to-end VLA of the past, which unified perception, understanding, and action. The current stage introduces enhanced VLA, integrating world models for "pre-action prediction." The future stage will mark a shift toward a brain-inspired mechanism.
AI² Robotics has already established a semi-automated production line capable of producing over 2,000 robots annually. Looking ahead, the company plans to build a new facility with a capacity of 20,000 to 30,000 units in the second half of this year. This facility aims to enable on-device deployment of large models.
