Li Auto Unveils Next-Gen Autonomous Driving Foundation Model MindVLA-o1: "See Better, Think Deeper"

At the NVIDIA GTC 2026 conference, Zhan Kun, Head of Foundation Models at Li Auto, officially unveiled MindVLA-o1, the company’s next-generation autonomous driving foundation model. The model significantly enhances autonomous driving capabilities through five key technological innovations: 3D spatial understanding, multimodal reasoning, unified behavior generation, closed-loop reinforcement learning, and hardware-software co-design. MindVLA-o1 fuses camera and LiDAR data to achieve precise 3D perception; incorporates a latent world model to simulate future scenarios and improve decision-making foresight; employs a VLA-MoE architecture to generate smooth, physically plausible trajectories; leverages a proprietary world simulator to dramatically accelerate model training and evolution; and, through hardware-software co-optimization, substantially shortens deployment cycles, enabling efficient on-device execution of large models on automotive-grade chips.