对话元戎启行CEO周光：VLA模型主要成本是AI芯片，已实现近10万辆辅助驾驶方案交付

Core Viewpoint - The launch of the DeepRoute IO 2.0 platform by Yuanrong Qixing marks a significant advancement in the field of autonomous driving, utilizing the innovative VLA (Vision-Language-Action) model to enhance safety and comfort in complex driving scenarios [2][6]. Group 1: Technology and Innovation - The VLA model integrates visual perception, semantic understanding, and action decision-making, representing a breakthrough compared to traditional end-to-end models [2]. - DeepRoute IO 2.0 is designed with a "multi-modal + multi-chip + multi-vehicle" adaptability, supporting both LiDAR and pure vision versions for various mainstream passenger car platforms [2][7]. - The VLA model's architecture allows for better generalization and adaptability to real-world driving conditions, overcoming the limitations of traditional models [7]. Group 2: Commercialization and Market Position - Yuanrong Qixing has secured over 10 model-specific collaborations and delivered nearly 100,000 vehicles equipped with urban navigation assistance systems, positioning itself in the industry's top tier [3][7]. - The company anticipates that by 2025, more than 200,000 vehicles featuring its combined assistance driving solutions will enter the consumer market [7]. - The company has completed six rounds of financing, raising over $500 million (approximately 3.57 billion RMB), with significant investments from major players like Alibaba and Fosun [7]. Group 3: Future Directions and Goals - The company aims to expand the application of the VLA model beyond automotive to robotics, indicating a vision for general artificial intelligence (AGI) in the physical world [3][4]. - Future developments will focus on enhancing safety in autonomous driving, with a commitment to defensive driving principles [8]. - The company plans to adopt a large model approach similar to Tesla's for developing L4 and L5 autonomous driving capabilities, emphasizing the need for a shift in the traditional definitions of autonomous driving [9].