随着自动驾驶技术的不断进步，理想汽车近日宣布其第三代自动驾驶技术架构已经完成内测，即将推向市场。这一架构采用了端到端 + VLM（视觉语言模型） + 世界模型的技术，旨在大幅提升智能驾驶能力。该技术架构已在千人范围内进行测试，结果显示，它能够提高AI的信息处理效率，增强对复杂路况的理解和应对能力，将智能驾驶行为拟人化。
理想汽车的这一技术突破，得益于生成式AI技术的兴起。目前，众多造车新势力都在探索视觉语言模型与世界模型的新方法，端到端智能驾驶技术成为共同的研究方向。理想汽车发布的第三代自动驾驶技术架构，受到了诺贝尔奖得主丹尼尔・卡尼曼快慢系统理论的启发，在自动驾驶领域模拟人类的思考和决策过程。
该架构中的“快系统”负责处理简单任务，而“慢系统”则负责解决复杂甚至未知的交通场景。理想汽车智能驾驶副总裁郎咸朋表示，目前理想的智能驾驶已经全面切入端到端 + 大模型方案，车辆能够理解复杂路况和交通规则。
理想汽车的研究人员应用最近生成式AI领域兴起的视觉语言模型（VLM），在视觉理解和推理方面表现出了非凡的能力。这是业界首个提出自动驾驶快慢系统的工作，其方法充分结合了主流自动驾驶pipeline和具备逻辑思考的大模型pipeline。
通过这种类似无限的环境生成，理想可以对其自动驾驶系统进行充分的检验以及迭代。截至今年6月，理想的整体保有量已经超过80万，随着技术的不断成熟和市场接受度的提高，未来智能驾驶将给人们带来更加安全、便捷的出行体验。
英语如下：
News Title: “Li Auto: Plans for a New Breakthrough in L3 Autonomous Driving in the First Half of Next Year”
Keywords: Li L3, AI Driving, Collaborative Decision-Making
News Content:
As autonomous driving technology continues to advance, Li Auto recently announced that its third-generation autonomous driving technology architecture has completed internal testing and is poised for market release. This architecture employs a combination of end-to-end + VLM (Visual Language Model) + World Model technologies, aimed at significantly enhancing intelligent driving capabilities. The technology has been tested on a thousand-person scale, with results indicating that it can improve the efficiency of AI information processing, enhance the understanding and response to complex road conditions, and anthropomorphize intelligent driving behaviors.
Li Auto’s technological breakthrough is attributed to the rise of generative AI technologies. Currently, many new automotive forces are exploring new methods of integrating visual language models with world models, with end-to-end intelligent driving technology becoming a common research direction. The third-generation autonomous driving technology architecture released by Li Auto is inspired by the Nobel laureate Daniel Kahneman’s fast and slow system theory, simulating the human thought and decision-making process in the autonomous driving field.
The “fast system” within the architecture is responsible for handling simple tasks, while the “slow system” deals with complex or unknown traffic scenarios. Li Auto’s Vice President of Intelligent Driving, Lang Xianpeng, stated that the company’s intelligent driving technology has fully entered the end-to-end + large model solution, allowing vehicles to understand complex road conditions and traffic rules.
Li Auto’s researchers have applied the emerging visual language model (VLM) in the generative AI field to demonstrate exceptional abilities in visual understanding and reasoning. This is the first industry proposal for an autonomous driving fast and slow system, which combines mainstream autonomous driving pipelines with pipelines that include logical thinking large models.
Through this approach similar to infinite environment generation, Li Auto can thoroughly test and iterate its autonomous driving system. As of June this year, the overall fleet of Li Auto has exceeded 800,000 vehicles. With the continuous maturity of technology and increasing market acceptance, future intelligent driving will bring people a safer and more convenient travel experience.
【来源】https://www.jiqizhixin.com/articles/2024-08-06-5
Views: 1