地平线吕鹏:端到端是基石,做不好端到端就做不好VLA
2 1 Shi Ji Jing Ji Bao Dao·2025-12-22 13:23

Core Viewpoint - The article emphasizes the importance of end-to-end technology in the development of autonomous driving solutions, highlighting Horizon's commitment to this approach as a foundation for future advancements in the industry. Market Overview - In the first three quarters of this year, the market share for passenger cars priced above 200,000 yuan accounted for 30%, while those below 130,000 yuan reached 50%, with many lower-priced models lacking urban auxiliary driving features [1]. - This gap in the market is attracting companies like Horizon and Momenta to accelerate their strategies to capture market opportunities [1]. Product Development - Horizon launched its Horizon SuperDrive (HSD) solution based on the Journey 6 series chips in April, entering mass production by November with the launch of the Exeed ET5 and Deep Blue L06 models, achieving over 12,000 activations within two weeks [1][2]. - The company aims to make urban auxiliary driving features available in vehicles priced around 100,000 yuan, targeting a production scale of ten million units in the next 3-5 years [2]. Technological Strategy - Horizon is one of the few companies firmly committed to the end-to-end approach in autonomous driving, believing that a solid end-to-end foundation is essential for integrating new modalities and enhancing product performance [3][7]. - The company has invested 90% of its R&D resources into developing and implementing end-to-end technology since the end of 2024 [2]. Technical Insights - Horizon's end-to-end system is described as a complete solution, contrasting with two-stage systems that may lose information during processing [4][5]. - The company believes that a robust end-to-end model is crucial for achieving high performance and seamless driving experiences, akin to human driving instincts [6][9]. Future Directions - Horizon's future plans include enhancing its end-to-end technology while exploring the integration of world models and reinforcement learning as auxiliary components to improve overall system performance [9][10]. - The focus remains on product experience and safety, with an emphasis on market acceptance rather than getting caught up in new terminologies or concepts [9].