智能化驾驶安全性和技术路径再度掀起市场热议。在《智能网联汽车组合辅助驾驶系统安全要求》强制性国标（下称“L2强标”）征求意见中，激光雷达首次纳入强标体系中，而随后被召回的超11万辆小米标准版SU7(配置|询价)均未配备激光雷达。激光雷达与纯视觉两大技术路径再次掀起市场热议。

　　长期以来，两大技术阵营在成本、安全、反应时间等层面争议颇多，伴随今年智能化进程加速，国内车企、自动驾驶方案解决商等相关方在两大路径的选择、分歧更为明显。在视觉一派，特斯拉、小鹏等厂商已明确通过高算力芯片、高精度摄像头提升自动驾驶技术；在激光雷达等多传感器融合一派，华为、速腾聚创等仍坚持搭载激光雷达提高安全冗余等。

　　关注点一：激光雷达、纯视觉两种路径，是否都能满足强标提出的多场景安全？

　　针对强标提出的多场景安全要求，两派从业者均向第一财经提及，在L2级别的辅助驾驶下，两种技术路线都可以满足L2强标，但具体执行细则上有所区别。

　　速腾聚创市场总监谢阗地向第一财经表示，此次L2强标不仅明确规定了激光雷达作为新型感知单元，有硬件一致性和失效检测的要求，还从测试的角度给出了指引，通过夜间及隧道等复杂光线环境下的测试，肯定了激光雷达对组合驾驶辅助功能的积极作用。

　　速腾聚创认为，在障碍物探测方面，以在高快速路上识别路面褐色纸箱避免碰撞为例，面对体积小、颜色深、RCS低的褐色纸箱时，毫米波雷达和摄像头难以在较远距离准备识别，激光雷达则能有效规避上述局限。

　　另有激光雷达从业者表示，激光雷达和L2强标的关系类似文具和考卷的关系，激光雷达能够在技术上更快速地达到强标的要求，完全抛弃激光雷达的解决方案可能会耽误车企在其他领域的进度，从而产生更高的成本，对车企整体收益不见得是好事。

　　而有视觉方案从业者则认为，依靠迭代进化的AI模型、视觉感知和本地端高算力等结合上车同样能够符合强标要求，激光雷达是视觉的补充。该人士称：“现阶段，我们可以通过毫米波雷达、超声波雷达等其他传感器搭配摄像头同步感知。”他还表示，在夜间环境下，搭载了激光雷达的车辆和纯视觉的车辆识别能力差不多，前者一样“看”不清。

　　关注点二：激光雷达、纯视觉两种路径，谁的安全性更高？

　　近期，特斯拉CEO马斯克再次提到，激光雷达等雷达会由于传感器间的争用降低驾驶的安全性。如果激光雷达与摄像头数据不一致，该听哪一方的？这种传感器的不确定性导致风险增加而降低安全。

　　另有纯视觉方案从业者告诉第一财经：“几年前算力和模型没有到达今天的水平时，激光雷达的信号比纯视觉更容易被算法处理。但是当算力和模型发展到现在，激光雷达出现的低延时问题，只有高帧率的纯视觉可以解决。”

　　在小鹏汽车董事长何小鹏的观点中，目前摄像头能力大幅提升，清晰度超过人眼，只要车端算力充足，视觉传感方案的响应速度可以比包含激光雷达的融合方案快数倍到10倍。去年，小鹏汽车战略转向了“纯视觉路线”，今年7月底，小鹏再次重申将坚持纯视觉技术路线，否认将重新使用激光雷达路线。但值得一提的是，小鹏在摄像头外，还搭载了毫米波雷达等其他类型的雷达，进一步提升安全冗余。

　　除了小鹏以外，也有不少厂商着手在两派中间进行转换。比如，元戎启行CEO周光认为，短期来看，激光雷达对通用障碍物识别仍有重要作用，大模型的知识库能力可以识别很多未知障碍物。但长期来看，随着大模型技术发展，视觉会在感知中扮演越来越重要的角色，大模型有望解决现在激光雷达的部分任务。

　　而激光雷达等多传感器融合方则认为，尽管激光雷达的数量和价格影响着汽车产品的售价，但却是当前提升辅助驾驶安全冗余最有效且直接的方法。

　　华为智能驾驶解决方案产品线总监李文广此前接受媒体采访时提到，面对晴转雨、团雾等突发驾驶场景，目前辅助驾驶的能力是不够的。他表示：“这不是因为‘脑子’不够聪明，而是‘看’不见。所以我们需要考虑一些能够在这些场景下发挥作用的传感器，比如分布式雷达，因为192线这类高线数的激光雷达识别这些场景也还是不够用的。”

　　小马智行CEO彭军认为，L4级别纯视觉很难实现完全安全。他告诉第一财经：“早年大家认为激光雷达价格昂贵，不适合用在车上，但现在情况已完全不同，激光雷达的价格基本与摄像头处于相同档位。”

　　综合两派的观点，当前，中国激光雷达企业占据全球超90%的汽车激光雷达市场份额，但纯视觉需要的高算力芯片厂商仍以英伟达等国外厂商居多，存在一定的“卡脖子”难题。

　　基于此，对纯视觉路线而言，不少主机厂或Tier 1也已经布局研发高算力芯片，这也将成为纯视觉路径未来提升自动驾驶能力的新研发方向。

　　激光雷达价格近年来也在不断下探，比如禾赛科技面向ADAS（高级驾驶辅助系统）量产应用的远距激光雷达产品在实现大规模出货的前提下，售价或能减半降至200美元。

　　而搭载激光雷达的L2级别新车也层出不穷。根据佐思汽研发布的《中国汽车传感器技术与数据趋势月度监测报告（2025年6月版）》，L3以下的智能辅助驾驶车型销量超66.7万辆，其中60%的车型都搭载了激光雷达。今年1~4月，激光雷达安装量超过48万颗，同比增幅超50%，其中自主品牌安装量超28万颗。但在纯视觉路径的不断发展过程中，二者谁能赢得未来更多厂商青睐，仍有待观望。

The safety and technological path of intelligent driving have once again sparked market discussions. In the solicitation of opinions on the mandatory national standard "Safety Requirements for Intelligent Connected Vehicle Combination Assisted Driving Systems" (hereinafter referred to as the "L2 Strong Standard"), LiDAR was included in the Strong Standard System for the first time, and over 110000 Xiaomi Standard Edition SU7 (configuration | inquiry) that were subsequently recalled were not equipped with LiDAR. The two major technological paths of LiDAR and pure vision have once again sparked market discussions.

For a long time, there have been many controversies between the two major technology camps in terms of cost, safety, and response time. With the acceleration of the intelligent process this year, the choices and differences between domestic car companies, autonomous driving solution providers, and other relevant parties in the two paths have become more apparent. On the visual side, manufacturers such as Tesla and Xiaopeng have explicitly improved their autonomous driving technology through high computing power chips and high-precision cameras; In the field of multi-sensor fusion such as LiDAR, Huawei, Sagitar Juchuang, and others still insist on using LiDAR to improve safety redundancy.

Focus 1: Can both LiDAR and pure vision paths meet the multi scene safety requirements proposed by Strong Standards?

In response to the multi scenario safety requirements proposed by the strong standard, both practitioners have mentioned to First Financial that under L2 level assisted driving, both technical routes can meet the L2 strong standard, but there are differences in the specific implementation details.

Xie Tiandi, Marketing Director of Sagitar Juchuang, told First Financial that the L2 strong standard not only clearly stipulates the requirement of hardware consistency and failure detection for laser radar as a new sensing unit, but also provides guidance from a testing perspective. Through testing in complex light environments such as night and tunnels, the positive role of laser radar in combined driving assistance functions has been affirmed.

Sagitar Juchuang believes that in obstacle detection, taking the identification of brown cardboard boxes on highways to avoid collisions as an example, when facing small, dark colored, and low RCS brown cardboard boxes, millimeter wave radar and cameras are difficult to prepare for recognition at a long distance, while laser radar can effectively avoid the above limitations.

Other LiDAR practitioners have expressed that the relationship between LiDAR and L2 strong standards is similar to that between stationery and exam papers. Lidar can meet the requirements of strong standards more quickly in technology, and completely abandoning LiDAR solutions may delay the progress of car companies in other fields, resulting in higher costs, which may not be a good thing for the overall revenue of car companies.

However, some visual solution practitioners believe that combining iterative evolution of AI models, visual perception, and local high computing power can also meet the requirements of strong standards, and LiDAR is a visual supplement. The person said, "At present, we can use other sensors such as millimeter wave radar and ultrasonic radar to synchronize perception with cameras." He also stated that in nighttime environments, vehicles equipped with LiDAR have similar recognition capabilities to pure visual vehicles, and the former cannot be seen clearly.

Focus 2: Which of the two paths, LiDAR and pure vision, has higher safety?

Recently, Tesla CEO Musk once again mentioned that radars such as LiDAR can reduce driving safety due to competition between sensors. If the LiDAR and camera data are inconsistent, which side should I listen to? The uncertainty of this type of sensor increases risk and reduces safety.

Another pure vision solution practitioner told First Financial, "A few years ago, when computing power and models did not reach today's level, LiDAR signals were more easily processed by algorithms than pure vision. However, when computing power and models have developed to this point, the low latency problem of LiDAR can only be solved by high frame rate pure vision

In the view of He Xiaopeng, Chairman of Xiaopeng Motors, the current camera capability has greatly improved, with clarity exceeding that of the human eye. As long as the computing power on the vehicle side is sufficient, the response speed of the visual sensing solution can be several to ten times faster than the fusion solution containing LiDAR. Last year, Xiaopeng Motors shifted its strategy towards a "pure visual route". At the end of July this year, Xiaopeng reiterated its commitment to the pure visual technology route and denied using the LiDAR route again. But it is worth mentioning that Xiaopeng is equipped with other types of radar such as millimeter wave radar in addition to the camera, further enhancing safety redundancy.

In addition to Xiaopeng, many manufacturers have also started to switch between the two factions. For example, Zhou Guang, CEO of Yuanrong Qixing, believes that in the short term, LiDAR still plays an important role in general obstacle recognition, and the knowledge base capability of large models can identify many unknown obstacles. But in the long run, with the development of big model technology, vision will play an increasingly important role in perception, and big models are expected to solve some of the tasks of current LiDAR.

However, multi-sensor fusion parties such as LiDAR believe that although the quantity and price of LiDAR affect the selling price of automotive products, it is currently the most effective and direct method to improve the safety redundancy of assisted driving.

Li Wenguang, the product line director of Huawei's intelligent driving solutions, previously mentioned in a media interview that the current ability of assisted driving is insufficient in the face of sudden driving scenarios such as sunny to rainy and foggy conditions. He said, "This is not because the 'brain' is not smart enough, but because it cannot see. So we need to consider some sensors that can work in these scenarios, such as distributed radar, because high line count LIDAR like 192 lines are still not enough to recognize these scenarios

Peng Jun, CEO of Xiaoma Zhixing, believes that it is difficult to achieve complete safety with L4 level pure vision. He told First Financial, "In the early days, people thought that LiDAR was expensive and not suitable for use in cars, but now the situation is completely different. The price of LiDAR is basically in the same range as that of cameras

Taking into account the views of both factions, currently, Chinese LiDAR companies occupy over 90% of the global automotive LiDAR market share. However, foreign manufacturers such as Nvidia still dominate the high computing power chip manufacturers required for pure vision, which poses a certain bottleneck problem.

Based on this, for the pure visual route, many host manufacturers or Tier 1 have already laid out the research and development of high computing power chips, which will also become a new research and development direction for the pure visual route to enhance autonomous driving capabilities in the future.

In recent years, the price of LiDAR has been continuously decreasing. For example, the long-range LiDAR products developed by Hesai Technology for ADAS (Advanced Driver Assistance Systems) mass production applications may be halved to $200 on the premise of achieving large-scale shipments.

And L2 level new cars equipped with LiDAR are also emerging one after another. According to the "Monthly Monitoring Report on China's Automotive Sensor Technology and Data Trends (June 2025 Edition)" released by Zosi Automotive Research, sales of intelligent assisted driving models below L3 have exceeded 667000 units, of which 60% of the models are equipped with LiDAR. From January to April this year, the installation of LiDAR exceeded 480000 units, with a year-on-year increase of over 50%. Among them, the installation of domestic brands exceeded 280000 units. However, in the continuous development of the pure visual path, it remains to be seen which of the two can win the favor of more manufacturers in the future.