近日，Momenta首席执行官曹旭东在CCTV财经《对话》中介绍到，现在技术已经进化到第六代了，称 为"强化学习的大模型"。 他提到，第五代是一个模仿学习，就是模仿人，能力的上限可能是接近人、达到人的水平。"有点像从 小学到中学到大学一路在学习，都是有老师的指导的，但真的要超过老师很难。而强化学习是在实践中 探索，成功了有奖励，失败了有惩罚。这样就能够探索出来更好的驾驶行为，有机会达到人类水平或者 超过人类水平。" 曹旭东表示，现在公司已经量产了50多万台车了。"这是什么概念呢？1000万台车每年驾驶的里程会达 到千亿公里，而人类一辈子可能只达到百万公里，那就10万倍的人类经验。它可以在云端的训练环境里 面遇到10万次，可能第一次的时候手忙脚乱，等到1000次、1万次的时候就已经非常老司机了。" "等到10万次的时候，它已经学会直觉驾驶了，知道这种挑战场景最优的驾驶策略是什么，能够实现最 安全、最高效的驾驶。"他说。 新浪声明：所有会议实录均为现场速记整理，未经演讲者审阅，新浪网登载此文出于传递更多信息之目 的，并不意味着赞同其观点或证实其描述。 责任编辑：李昂 近日，Momenta首席执行官曹旭东在CCTV ...

Core Insights - The article highlights the transition of China's intelligent driving industry from concept to practical application, with Chery's commitment to its intelligent driving strategy serving as a milestone [1][3]. Industry Overview - By 2025, the Chinese intelligent driving industry is expected to shift from "parameter competition" to "real-world validation," with consumer expectations evolving from "availability" to "usability" and "reliability" [3]. - The current stage of the industry is characterized by both technological breakthroughs and challenges in implementation [4]. Chery's Commitment - Chery's chairman publicly committed to equipping all models with the Falcon intelligent driving assistance system within the year, a move that sparked industry discussions due to the previous trend of high-level intelligent driving features being limited to premium models [3][6]. - As of the end of the year, Chery successfully integrated the Falcon system across all models, demonstrating its technical capabilities through real-world testing in complex driving conditions [3][6]. Challenges in Intelligent Driving - Many automakers face issues such as "feature reduction," "delayed functionality," and limitations to high-end models when delivering intelligent driving features [5]. - Current intelligent driving systems exhibit significantly higher error rates on unstructured roads compared to structured ones, with failure rates being 3-5 times higher [5]. Technical Foundation of Falcon Intelligent Driving - The Falcon system's success is attributed to a collaborative foundation of data, algorithms, and hardware, creating a "data loop - algorithm breakthrough - hardware redundancy" structure [7]. - Chery's Tianqiong Intelligent Computing Center has accumulated over 24 billion kilometers of driving assistance data, enhancing the system's adaptability across various road conditions [7][10]. Algorithm and Hardware Integration - The Falcon system utilizes the Momenta R6 reinforcement learning model, which allows for rapid decision-making in unforeseen scenarios, enhancing its performance in complex environments [10][11]. - The hardware setup includes a combination of sensors, ensuring reliable perception in challenging conditions, while the system's computational power is optimized for efficient data processing [12][14]. Long-term Strategy and Collaboration - Chery's approach to intelligent driving is rooted in a long-term commitment to technology development, having invested in intelligent technology since 2010 [17][19]. - The company employs a collaborative ecosystem model, partnering with various tech firms to enhance its capabilities while maintaining core technology independence [19]. Future Outlook - Chery aims to achieve end-to-end integration of its intelligent driving system by 2026, with ongoing updates to enhance functionality [21]. - The intelligent driving industry is moving towards a phase of "refined cultivation," focusing on real-world validation and user-centric solutions [22].