[{"data":1,"prerenderedAt":215},["ShallowReactive",2],{"blog-post-autonomous-decisions-the-bias-variance-tradeoff-in-self-driving-technology":3,"blog-featured-autonomous-decisions-the-bias-variance-tradeoff-in-self-driving-technology":24,"header-links":59,"corporate-footer":146,"footer-year":214},[4],{"id":5,"type":6,"lang":7,"primary_tag":8,"url":9,"title":10,"short_title":11,"subtitle":11,"description":12,"meta_description":12,"image":13,"img_alt":14,"author_name":15,"download_doc":11,"download_title":11,"download_btn":11,"webinar_video":11,"thumbnail":11,"author_id":11,"content":16,"publish_date":17,"featured":18,"is_gated":19,"is_infographic":11,"is_hidden":19,"is_active":20,"tags":21,"updated_at":23},261,"blog","en","Amnon Shashua","autonomous-decisions-the-bias-variance-tradeoff-in-self-driving-technology","自主决策：自动驾驶技术中的偏差-方差权衡",null,"大语言模型和自动驾驶应用中，单一人工智能系统与复合人工智能系统的比较","https:\u002F\u002Fstatic.mobileye.com\u002Fwebsite\u002Fus\u002Fcorporate\u002Fimages\u002Ff8167f28d4d800bd54b3e04c9fda5a5d_1715717822589.jpg","Monolithic versus compound AI systems in LLMs and autonomous driving.","Prof. Amnon Shashua and Prof. Shai Shalev-Shwartz","\u003Cp>Back in November 2022, the release of ChatGPT garnered widespread attention, not only for its versatility but also for its end-to-end design. This design involved a single foundational component, the GPT 3.5 large language model, which was enhanced through both supervised learning and reinforcement learning from human feedback to support conversational tasks. This holistic approach to AI was highlighted again with the launch of Tesla’s latest FSD system, described as an end-to-end neural network that processes visual data directly “from photons to driving control decisions,\" without intermediary steps or “glue code.\" \u003C\u002Fp>\n\u003Cp>While AI models continue to evolve, the latest generation of ChatGPT has moved away from the monolithic E2E approach. Consider the following example of a conversation with ChatGPT:\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Fstatic.mobileye.com\u002Fwebsite\u002Fus\u002Fcorporate\u002Fimages\u002Fchatgpt_blog2.png\" alt=\"\" width=\"746\" height=\"448\" \u002F>\u003C\u002Fp>\n\u003Cp>When asked to compute \"what is 3456 * 3678?,\" the system first translates the question into a short Python script to perform the calculation, and then formats the output of the script into a coherent natural language text. This demonstrates that ChatGPT does not rely on a single, unified process. Instead, it integrates multiple subsystems—including a robust deep learning model (GPT LLM) and separately coded modules. Each subsystem has its defined role, interfaces, and development strategies, all engineered by humans. Additionally, 'glue code' is employed to facilitate communication between these subsystems. This architecture is referred to as “\u003Ca href=\"https:\u002F\u002Fbair.berkeley.edu\u002Fblog\u002F2024\u002F02\u002F18\u002Fcompound-ai-systems\u002F\" target=\"_blank\" rel=\"noopener\">Compound AI Systems\u003C\u002Fa>\u003Ca href=\"..\u002F..\u002Fblog\u002Fautonomous-decisions-the-bias-variance-tradeoff-in-self-driving-technology\u002F#_edn1\">\u003Csup>1\u003C\u002Fsup>\u003C\u002Fa>” (CAIS).\u003C\u002Fp>\n\u003Cp>Before we proceed, it is crucial to dispel misconceptions about which system architecture is \"new\" or \"traditional\". Despite the hype, the E2E approach in autonomous driving is not a novel concept; it dates back to \u003Ca href=\"https:\u002F\u002Fproceedings.neurips.cc\u002Fpaper\u002F1988\u002Ffile\u002F812b4ba287f5ee0bc9d43bbf5bbe87fb-Paper.pdf\" target=\"_blank\" rel=\"noopener\">the Alvinn project (Pomeraleau, 1989)\u003C\u002Fa>\u003Ca href=\"..\u002F..\u002Fblog\u002Fautonomous-decisions-the-bias-variance-tradeoff-in-self-driving-technology\u002F#_edn2\">\u003Csup>2\u003C\u002Fsup>\u003C\u002Fa>. The CAIS approach is also not new, but as we have shown above, it has been adopted by the most recent versions of ChatGPT.\u003C\u002Fp>\n\u003Cp>\u003Cem>This blog aims to explore the nuances between E2E systems and CAIS, by drawing a deep connection to the \u003C\u002Fem>\u003Ca href=\"https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FBias–variance_tradeoff\" target=\"_blank\" rel=\"noopener\">\u003Cem>bias-variance tradeoff\u003C\u002Fem>\u003C\u002Fa>\u003Ca href=\"..\u002F..\u002Fblog\u002Fautonomous-decisions-the-bias-variance-tradeoff-in-self-driving-technology\u002F#_edn3\">\u003Cem>\u003Csup>3\u003C\u002Fsup>\u003C\u002Fem>\u003C\u002Fa>\u003Cem> in machine learning and statistics. For concreteness, we focus the discussion on self-driving systems, but the connection is applicable more generally to the design of any AI-based system.\u003C\u002Fem>\u003C\u002Fp>\n\u003Ch3>\u003Cstrong>The bias-variance tradeoff\u003C\u002Fstrong>\u003C\u002Fh3>\n\u003Cp>The grand question, driving any school of thought for building a data-driven system, is the \"bias\u002Fvariance\" tradeoff. Bias, also known as “approximation error,\" means that our learning system cannot reflect the full richness of reality. Variance, also known as “generalization error,\" means that our learning system overfits to the observed data, and fails to generalize to unseen examples.\u003C\u002Fp>\n\u003Cp>The total error of the learned model is the sum of the approximation and generalization errors, so in order to reach a sufficiently small error, we need to delicately control both. There is a tradeoff between the two terms since we can decrease the generalization error by restricting the learned model to come from a specific family of models, but this might introduce a bias if the chosen family of models cannot reflect the full richness of reality.\u003C\u002Fp>\n\u003Cp>Based on this background, we can formalize the two approaches for building a self-driving system. A CAIS, or an “engineered system,\" deliberately puts architectural restrictions on the self-driving system for the sake of reducing the generalization error. This introduces some bias. For example, going from \"photons\" to a “sensing state,\" which is a model of reality surrounding the host vehicle—location and measurements of road users, roadway structures, drivable paths, obstacles and so forth—and from there to control decisions introduces bias. The reason for this bias is that the sensing state might not be rich enough to reflect reality to its fullest and therefore the capacity of the system is constrained. In contrast, an E2E network skipping the sensing state step and instead mapping incoming videos directly to vehicle control decisions would not suffer from this bias of the “sensing state abstraction.\" On the other hand, the E2E approach will have a higher generalization error, and the approach advocated by the E2E proponents is to compensate for this error with huge amounts of data, which in turn necessitates a huge investment in compute, storage, and data engines.\u003C\u002Fp>\n\u003Cp>To recap, the E2E approach – in the simplistic form being communicated to the public – is to define a system with zero bias while incrementally reducing variance through volumes of data for training the system with the purpose of gradually eliminating all “corner cases\". In an engineered approach, on the other hand, the system starts with a built-in bias due to the abstraction of the sensing state and driving policy while (further) reducing variance through data fed into separate subsystems with a high-level fusion glue-code. The amount of data required for each subsystem is exponentially smaller than the amount of data required for a single monolithic system.\u003C\u002Fp>\n\u003Ch3>\u003Cstrong>The devil is in the AI details\u003C\u002Fstrong>\u003C\u002Fh3>\n\u003Cp>The story, however, is more delicate than the above dichotomy. Starting from the bias element, it’s not that E2E systems will have zero bias—since the neural network resides on an on-board computer in the car, its size is constrained by the available compute and memory. Therefore, the limited compute available in a car introduces bias. In addition, the approximation error of a well-engineered approach is not necessarily excessively large for several reasons. For one, Waymo is clear evidence that the bias of an engineered system is sufficiently small for building a safe autonomous car. Moreover, in a well-engineered system we can add a subsystem that skips some of the abstractions (e.g., the sensing state abstraction) and thus further reduce the bias in the overall system.\u003C\u002Fp>\n\u003Cp>The variance element is also more nuanced. In an \"engineered\" system, the variance is reduced through abstractions (such as sensing state) as well as through the high-level fusion of multiple subsystems. In a pure E2E system, the variance should be reduced only through more and more data. But this process of reducing variance through a data pipeline deserves more scrutiny. Take the notion of mean time between failures (MTBF) and let's assume that failures are measured by critical interventions. Let's take MTBF as a measure of readiness of a self-driving vehicle to operate in an \"eyes-off\" manner. In Mobileye's engagement with car makers the MTBF target is 10\u003Csup>7\u003C\u002Fsup> hours of driving. Just for reference, \u003Ca href=\"https:\u002F\u002Fwww.teslafsdtracker.com\u002Fhome\" target=\"_blank\" rel=\"noopener\">public data\u003C\u002Fa>\u003Ca href=\"..\u002F..\u002Fblog\u002Fautonomous-decisions-the-bias-variance-tradeoff-in-self-driving-technology\u002F#_edn4\">\u003Csup>4\u003C\u002Fsup>\u003C\u002Fa> on Tesla's recent V12.3.6 version of FSD stands around 300 miles per critical intervention which amounts to an MTBF of roughly 10 hours – which is 6 orders of magnitude away from the target MTBF. Let's assume that somehow the MTBF has reached 10\u003Csup>6\u003C\u002Fsup> hours and we wish merely to improve it by one order of magnitude in order to reach 10\u003Csup>7\u003C\u002Fsup> hours. How much data would we need to collect? This is a question about the nature of the long tail of a distribution. To make things concrete, assume we have 1 million vehicles on the road driving one hour per day and the data in question is event-driven—i.e., when an intervention of the human driver occurs then a recording of some time around the event is being made and sent to the car maker for further training. An intervention event represents a \"corner case\" and the question of the long tail is how those corner cases are distributed. Consider the following scenario where B is the set of “bad” corner cases. An example of a heavy tail distribution is when B = {b_1,...,b_{1000}} and the probability of b_i to occur is 10\u003Csup>-9\u003C\u002Fsup> for every i. Even if we assume that when a corner case is discovered then we can somehow retrain the network and fix that corner case, without creating any new corner cases, we must knock off around 900 corner cases so that P(B) = 10\u003Csup>-7\u003C\u002Fsup>. Because the MTBF is 10\u003Csup>6\u003C\u002Fsup> then we encounter one corner case per day. It follows that we will need around 3 years to get this done. The point here is that no one knows how the long tail is structured - we gave one possible long tail scenario but in reality it could be worse or it could be better. As mentioned previously, while there is a precedent that the bias of an engineered system is sufficient for building a safe autonomous car (e.g. Waymo), there is still no precedent that reducing variance solely by a recurring data engine is sufficient for building a safe autonomous car.\u003C\u002Fp>\n\u003Cp>It follows that to solely double-down on a data pipeline might be too risky. What else can be done in the E2E approach for reducing variance? To lead into it lets ask ourselves:\u003C\u002Fp>\n\u003Cp>\u003Cbr \u002F>  (i) Why does Tesla FSD have a sensing state in their display? The idea of an E2E system is you go from \"photons to control\" while skipping the need to build a sensing state. Are they doing that solely for the purpose of notifying the driver what the system “sees?\" Or does it have a more tacit purpose?\u003C\u002Fp>\n\u003Cp>\u003Cbr \u002F>  (ii) Why has Tesla \u003Ca href=\"https:\u002F\u002Fwww.theverge.com\u002F2024\u002F5\u002F7\u002F24151497\u002Ftesla-lidar-bought-luminar-elon-musk-sensor-autonomous\" target=\"_blank\" rel=\"noopener\">purchased 2,000 lidars\u003C\u002Fa>\u003Ca href=\"..\u002F..\u002Fblog\u002Fautonomous-decisions-the-bias-variance-tradeoff-in-self-driving-technology\u002F#_edn5\">\u003Csup>5\u003C\u002Fsup>\u003C\u002Fa> from Luminar? Presumably for creating ground truth (GT) data for a supervised training. But why?\u003C\u002Fp>\n\u003Cp>The two riddles are of course related.\u003C\u002Fp>\n\u003Cp>Imagine an E2E network comprising of a backbone and two heads - one for outputting vehicle control and the other for outputting the sensing state. Such a network is still technically E2E (from photons to control) but also has a branch for sensing state. The sensing state branch needs to be trained in a supervised manner from GT data, hence the need for 2,000 lidars. The real question is whether the GT data can be created automatically without manual labeling. The answer is definitely yes because Mobileye does that. We have an \"auto-GT\" pipeline for training for sensing state. And, the reason why you would want a branch outputting the sensing state is not merely for displaying the sensing state to the driver. The real purpose of the sensing state branch is to reduce the variance (and more importantly the sample complexity of the system which is the amount of data needed for training) of the system through the \"multi-tasking\" principle. We addressed this \u003Ca href=\"https:\u002F\u002Farxiv.org\u002Fpdf\u002F1604.06915\" target=\"_blank\" rel=\"noopener\">back in 2016\u003C\u002Fa>\u003Ca href=\"..\u002F..\u002Fblog\u002Fautonomous-decisions-the-bias-variance-tradeoff-in-self-driving-technology\u002F#_edn6\">\u003Csup>6\u003C\u002Fsup>\u003C\u002Fa> and gave as an example an agricultural vehicle on the side of the road. The probability of observing a rare vehicle type somewhere in the image is much higher than the probability of observing such a vehicle immediately in front of us. Therefore, without a sensing state head (that detects the rare vehicle type on the shoulder even if it is not affecting the control of the host vehicle) one would need much more data in order to see such a vehicle immediately in front us, which affects the control of the host vehicle. What this comes to show is that the sensing state abstraction is important as it hints to the neural network that a good approach for giving correct control commands may need to understand the concept of vehicles and to detect all of the vehicles around us. Importantly, this abstraction is learned through supervised learning, using GT data which is created automatically (by a well-engineered offline system).\u003C\u002Fp>\n\u003Cp>Another component of the popular E2E narrative is \"no glue-code\" in the system. No glue-code means no bugs being entered by careless engineers. But this too is a misconception. There is glue-code – not in the neural network but in the process of preparing the data for the E2E training. One clear example is the automatic creation of GT data. Elon Musk gave such an example – that of human drivers not respecting stop signs as they should and instead perform a \u003Ca href=\"https:\u002F\u002Fwww.teslarati.com\u002Ftesla-rolling-stop-markey-blumenthal-letter-elon-musk\u002F\" target=\"_blank\" rel=\"noopener\">rolling stop\u003C\u002Fa>\u003Ca href=\"..\u002F..\u002Fblog\u002Fautonomous-decisions-the-bias-variance-tradeoff-in-self-driving-technology\u002F#_edn7\">\u003Csup>7\u003C\u002Fsup>\u003C\u002Fa>. The rolling stop events had to be taken out of the training data so that the E2E system would not adopt bad behavior (as it is supposed to imitate humans). In other words, the glue-code (and bugs) are shifting from the system code to data curation code. Actually, it may be easier to detect bugs in system code (at least there are existing methodologies for that) than to detect bugs in data curation.\u003C\u002Fp>\n\u003Ch3>\u003Cstrong>Don’t get into the way of analytic solutions\u003C\u002Fstrong>\u003C\u002Fh3>\n\u003Cp>Finally, we would like to point out that when we have an analytical solution for a problem, it is certainly not better to use a purely E2E machine learning method. For example, the long multiplication exercise mentioned at the start of this blog highlights the fact that ChatGPT+, rightfully so, uses a good old calculator for the task and does not attempt to \"learn\" how to do long multiplication from a massive amount of data. In the driving policy stack (determining the actions the host vehicle should do and outputting the vehicle control commands) there are numerous analytical calculations. Knowing when to replace \"learning\" with analytical calculations is crucial for variance reduction but also for transparency, explainability and \u003Ca href=\"https:\u002F\u002Fwww.mobileye.com\u002Fopinion\u002Fmobileye-dxp-as-a-novel-approach\u002F\">tuning\u003C\u002Fa>\u003Ca href=\"..\u002F..\u002Fblog\u002Fautonomous-decisions-the-bias-variance-tradeoff-in-self-driving-technology\u002F#_edn8\">\u003Csup>8\u003C\u002Fsup>\u003C\u002Fa> (imitating humans is somewhat problematic because many of them are not good drivers).\u003C\u002Fp>\n\u003Ch3>\u003Cstrong>Final words\u003C\u002Fstrong>\u003C\u002Fh3>\n\u003Cp>AI is progressing at a remarkable pace. The revolution of foundation models of the like of ChatGPT began as a monolithic E2E model (back in 2022) whereas today it evolved into \u003Ca href=\"https:\u002F\u002Fopenai.com\u002Fchatgpt\u002Fpricing\" target=\"_blank\" rel=\"noopener\">ChatGPT+\u003C\u002Fa>\u003Ca href=\"..\u002F..\u002Fblog\u002Fautonomous-decisions-the-bias-variance-tradeoff-in-self-driving-technology\u002F#_edn9\">\u003Csup>9\u003C\u002Fsup>\u003C\u002Fa> which represents an engineered solution built on top of AI components including the base LLM, plugins for retrieval, code interpreter and image generation tools. There is much to be said about the claims that the future of AI is \u003Ca href=\"https:\u002F\u002Fbair.berkeley.edu\u002Fblog\u002F2024\u002F02\u002F18\u002Fcompound-ai-systems\u002F\" target=\"_blank\" rel=\"noopener\">shifting\u003C\u002Fa>\u003Ca href=\"..\u002F..\u002Fblog\u002Fautonomous-decisions-the-bias-variance-tradeoff-in-self-driving-technology\u002F#_edn10\">\u003Csup>10\u003C\u002Fsup>\u003C\u002Fa> from monolithic models to compound AI systems. We believe that this trend is even more important given the extremely high accuracy requirement for, and safety-critical aspect of, autonomous driving.\u003C\u002Fp>\n\u003Cp> \u003C\u002Fp>\n\u003Cp>\u003Cu>References:\u003C\u002Fu>\u003C\u002Fp>\n\u003Cp style=\"font-size: 1.3rem; margin-top: 1.4rem;\">\u003Ca id=\"_edn1\" href=\"#_ednref1\" name=\"_edn1\">\u003C\u002Fa>\u003Csup class=\"ref\">1\u003C\u002Fsup> Compound AI Systems\u003Cbr \u002F>\u003Ca href=\"https:\u002F\u002Fbair.berkeley.edu\u002Fblog\u002F2024\u002F02\u002F18\u002Fcompound-ai-systems\u002F\">https:\u002F\u002Fbair.berkeley.edu\u002Fblog\u002F2024\u002F02\u002F18\u002Fcompound-ai-systems\u002F\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp style=\"font-size: 1.3rem; margin-top: 1.4rem;\">\u003Ca id=\"_edn2\" href=\"#_ednref2\" name=\"_edn2\">\u003C\u002Fa>\u003Csup class=\"ref\">2\u003C\u002Fsup> the Alvinn project (Pomeraleau, 1989)\u003Cbr \u002F>\u003Ca style=\"word-wrap: break-word;\" href=\"https:\u002F\u002Fproceedings.neurips.cc\u002Fpaper\u002F1988\u002Ffile\u002F812b4ba287f5ee0bc9d43bbf5bbe87fb-Paper.pdf\">https:\u002F\u002Fproceedings.neurips.cc\u002Fpaper\u002F1988\u002Ffile\u002F812b4ba287f5ee0bc9d43bbf5bbe87fb-Paper.pdf\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp style=\"font-size: 1.3rem; margin-top: 1.4rem;\">\u003Ca id=\"_edn3\" href=\"#_ednref3\" name=\"_edn3\">\u003C\u002Fa>\u003Csup class=\"ref\">3\u003C\u002Fsup> Bias-variance tradeoff\u003Cbr \u002F>\u003Ca href=\"https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FBias–variance_tradeoff\">https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FBias–variance_tradeoff\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp style=\"font-size: 1.3rem; margin-top: 1.4rem;\">\u003Ca id=\"_edn4\" href=\"#_ednref4\" name=\"_edn4\">\u003C\u002Fa>\u003Csup class=\"ref\">4\u003C\u002Fsup> Public data on Tesla's recent V12.3.6 version of FSD\u003Cbr \u002F>\u003Ca href=\"https:\u002F\u002Fwww.teslafsdtracker.com\u002Fhome\">https:\u002F\u002Fwww.teslafsdtracker.com\u002Fhome\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp style=\"font-size: 1.3rem; margin-top: 1.4rem;\">\u003Ca id=\"_edn5\" href=\"#_ednref5\" name=\"_edn5\">\u003C\u002Fa>\u003Csup class=\"ref\">5\u003C\u002Fsup> Tesla purchased 2000 Lidars from Luminar\u003Cbr \u002F>\u003Ca href=\"https:\u002F\u002Fwww.theverge.com\u002F2024\u002F5\u002F7\u002F24151497\u002Ftesla-lidar-bought-luminar-elon-musk-sensor-autonomous\">https:\u002F\u002Fwww.theverge.com\u002F2024\u002F5\u002F7\u002F24151497\u002Ftesla-lidar-bought-luminar-elon-musk-sensor-autonomous\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp style=\"font-size: 1.3rem; margin-top: 1.4rem;\">\u003Ca id=\"_edn6\" href=\"#_ednref6\" name=\"_edn6\">\u003C\u002Fa>\u003Csup class=\"ref\">6\u003C\u002Fsup> On the Sample Complexity of End-to-end Training vs. Semantic Abstraction Training \u003Cbr \u002F>\u003Ca href=\"https:\u002F\u002Farxiv.org\u002Fpdf\u002F1604.06915\">https:\u002F\u002Farxiv.org\u002Fpdf\u002F1604.06915\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp style=\"font-size: 1.3rem; margin-top: 1.4rem;\">\u003Ca id=\"_edn7\" href=\"#_ednref7\" name=\"_edn7\">\u003C\u002Fa>\u003Csup class=\"ref\">7\u003C\u002Fsup> Human drivers not respecting stop signs as they should and instead perform a rolling stop\u003Cbr \u002F>\u003Ca href=\"https:\u002F\u002Fwww.teslarati.com\u002Ftesla-rolling-stop-markey-blumenthal-letter-elon-musk\u002F\">https:\u002F\u002Fwww.teslarati.com\u002Ftesla-rolling-stop-markey-blumenthal-letter-elon-musk\u002F\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp style=\"font-size: 1.3rem; margin-top: 1.4rem;\">\u003Ca id=\"_edn8\" href=\"#_ednref8\" name=\"_edn8\">\u003C\u002Fa>\u003Csup class=\"ref\">8\u003C\u002Fsup> Mobileye DXP as a novel approach \u003Cbr \u002F>\u003Ca href=\"https:\u002F\u002Fwww.mobileye.com\u002Fopinion\u002Fmobileye-dxp-as-a-novel-approach\u002F\">https:\u002F\u002Fwww.mobileye.com\u002Fopinion\u002Fmobileye-dxp-as-a-novel-approach\u002F\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp style=\"font-size: 1.3rem; margin-top: 1.4rem;\">\u003Ca id=\"_edn9\" href=\"#_ednref9\" name=\"_edn9\">\u003C\u002Fa>\u003Csup class=\"ref\">9\u003C\u002Fsup> ChatGPT+\u003Cbr \u002F>\u003Ca href=\"https:\u002F\u002Fopenai.com\u002Fchatgpt\u002Fpricing\">https:\u002F\u002Fopenai.com\u002Fchatgpt\u002Fpricing\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp style=\"font-size: 1.3rem; margin-top: 1.4rem;\">\u003Ca id=\"_edn10\" href=\"#_ednref10\" name=\"_edn10\">\u003C\u002Fa>\u003Csup class=\"ref\">10\u003C\u002Fsup> Future of AI is shifting from monolithic models to compound AI systems\u003Cbr \u002F>\u003Ca href=\"https:\u002F\u002Fbair.berkeley.edu\u002Fblog\u002F2024\u002F02\u002F18\u002Fcompound-ai-systems\u002F\">https:\u002F\u002Fbair.berkeley.edu\u002Fblog\u002F2024\u002F02\u002F18\u002Fcompound-ai-systems\u002F\u003C\u002Fa>\u003C\u002Fp>","2024-05-15T04:00:00.000Z",0,false,true,[8,22],"Autonomous Driving","2026-08-04T08:00:37.472Z",[25,37,48],{"id":26,"type":27,"lang":7,"primary_tag":22,"url":28,"title":29,"short_title":11,"subtitle":11,"description":30,"meta_description":30,"image":31,"img_alt":32,"author_name":11,"download_doc":11,"download_title":11,"download_btn":11,"webinar_video":11,"thumbnail":11,"author_id":11,"content":33,"publish_date":34,"featured":35,"is_gated":19,"is_infographic":11,"is_hidden":19,"is_active":20,"tags":36,"updated_at":23},329,"news","mobileye-to-establish-vertically-integrated-robotaxi-business","Mobileye 将打造垂直整合Robotaxi业务","全新业务布局突破原有自动驾驶系统供应商定位，与现有车企、移动出行合作项目形成互补","https:\u002F\u002Fstatic.mobileye.com\u002Fwebsite\u002Fus\u002Fcorporate\u002Fimages\u002F86fdfc495d3212c3c26221fbe4f90e66_1781548502451.webp","Mobileye 拟推出的Robotaxi服务概念示意图。","\u003Cp>\u003Cstrong>中国上海，2026年6月16日\u003C\u002Fstrong> — Mobileye 今日宣布，计划进一步拓展自动驾驶出租车（Robotaxi）业务版图，从提供自动驾驶技术延伸至自主运营自动驾驶网约车服务。该新项目计划于2027年率先落地美国一座城市，标志着Mobileye战略的重要升级。届时，Mobileye将整合行业领先自动驾驶技术、车队运营、乘客出行服务与出行管理能力，打造垂直整合解决方案。这一新业务是在Mobileye现有商业模式基础上的进一步拓展 —— Mobileye将继续作为自动驾驶技术供应商，为全球汽车制造商和出行服务提供商提供解决方案，在持续为客户部署提供支持的同时，增设全新运营板块。\u003C\u002Fp>\n\u003Cp>目前 Mobileye Drive 作为独立自动驾驶系统，正整合进合作伙伴项目。根据新举措的规划，Mobileye将打通Robotaxi价值链，依托Mobileye Drive系统，整合旗下Moovit出行平台及消费端应用、多模式出行规划、自动驾驶运营调度、车队管理技术以及远程协助基础设施。\u003C\u002Fp>\n\u003Cp>这一新举措不会改变Mobileye继续向汽车制造商、出行运营商及其他客户提供Mobileye Drive的承诺。公司将自营Robotaxi业务视作互补市场化路径，其将有助于加速自动驾驶落地部署，沉淀一线运营经验，并进一步验证Mobileye Drive平台的规模化应用能力。公司计划客户合作项目与自有运营车队双线同步推进。\u003C\u002Fp>\n\u003Cp>Mobileye规划首批部署约100台自动驾驶车辆，并于2027年开始在美国一座大都市投入运营。部署工作将分阶段推进，旨在验证完全无人驾驶条件下的整套运营模式。待首支车队稳定运营后，公司还计划大幅扩大业务规模，目标未来五年内将车队规模扩展至约17,000辆。\u003C\u002Fp>\n\u003Cp>Mobileye创始人兼首席执行官Amnon Shashua教授表示：“Robotaxi领域的变革才刚刚起步，其重塑全球出行方式的潜力仍在持续增长。随着自动驾驶出行赛道热度攀升，行业高度依赖为数不多的技术供应商与商业模式。而我们认为，可以开拓全新的发展路径，依托深厚的自动驾驶技术积淀、强大的产业合作生态、以及覆盖整个出行体系的成熟落地能力，构建差异化布局。 二十余年来，Mobileye持续深耕自动驾驶底层技术研发。如今我们迈出了新的重要一步：自研技术叠加自主运营，开创一套可实现财务层面规模化扩展、跨区域复制的全球化Robotaxi业务。我们计划于2026年底前在美国举办资本市场日活动，届时将分享更多有关商业化、技术及运营方面的详细信息。”\u003C\u002Fp>\n\u003Cp>为完善端到端自动驾驶平台，Mobileye将与自动驾驶整车平台厂商、车队运营商、车辆集成合作伙伴以及关键技术供应商展开合作。依托这套生态体系，Mobileye将通过统一的业务部门，统筹自有自动驾驶网约车服务的建设与运营。”\u003C\u002Fp>\n\u003Cp>Amnon Shashua教授补充道：“这一举措并非取代现有合作伙伴关系，而是在此基础上的延伸。我们将一如既往地通过Mobileye Drive赋能车企与出行服务商；与此同时，运营自有出行服务也有助于我们加快技术的市场采用，积累一手运营经验，充分释放自动驾驶出行的潜力。”\u003C\u002Fp>\n\u003Cp>Mobileye Drive沉淀25年以上的计算机视觉、地图、感知及自动驾驶技术研发经验，搭载最新复合人工智能架构，融合多类AI算法，配套严谨的安全框架，并通过工程化架构实现落地。全球已有超2.3亿台车辆搭载Mobileye技术，在业内积累了业内领先的真实道路经验。\u003C\u002Fp>\n\u003Cp>这一业务布局同时也拓展了Moovit原有的业务战略。Moovit致力于解决复杂城市环境下，用户查询、规划、使用公共交通服务的全流程需求。Moovit出行平台覆盖全球112个国家、3500余座城市，支持45种语言，服务超17亿用户。Moovit在消费者出行服务、多模式线路规划、用户运营和车队运营方面的专业知识，为公司在全球范围内扩展自动驾驶出行服务提供了关键基础。\u003C\u002Fp>\n\u003Cp>___________________________________\u003C\u002Fp>\n\u003Ch6>Mobileye (Nasdaq: MBLY) leads the mobility revolution with our autonomous driving and driver-assistance technologies, harnessing world-renowned expertise in artificial intelligence, computer vision and integrated software and hardware. Since our founding in 1999, Mobileye has enabled the global adoption of advanced driver-assistance systems that save countless lives and reduce crashes, while pioneering groundbreaking technologies such as REM™ crowdsourced road intelligence, Imaging Radar and Compound AI. These technologies drive the ADAS and AV fields towards the future of mobility – enabling self-driving vehicles and mobility solutions at scale, and powering industry-leading ADAS products. Through 2025, more than 230 million vehicles worldwide have been built with Mobileye’s EyeQ technology inside. In 2026, Mobileye acquired Mentee Robotics to pursue the future of physical AI and humanoid robots. Since 2022, Mobileye has been listed independently from Intel (Nasdaq: INTC), which retains majority ownership. For more information, visit \u003Ca href=\"https:\u002F\u002Fwww.mobileye.com\" target=\"_blank\" rel=\"noopener noreferrer\">https:\u002F\u002Fwww.mobileye.com\u003C\u002Fa>.\u003C\u002Fh6>\n\u003Ch6>“Mobileye,” the Mobileye logo and Mobileye product names are registered trademarks of Mobileye Global. All other marks are the property of their respective owners.\u003C\u002Fh6>\n\u003Ch6>Forward-Looking Statements\u003C\u002Fh6>\n\u003Ch6>This press release contains forward-looking statements within the meaning of the Private Securities Litigation Reform Act of 1995. Statements in this release that are not statements of historical fact are forward-looking statements and should be evaluated as such, including statements regarding Mobileye’s expansion into a full ownership of an autonomous ride-hailing business, and the impact and potential benefits to Mobileye’s business. These statements often include words such as “anticipate,” “expect,” “suggests,” “plan,” “believe,” “intend,” “estimates,” “targets,” “projects,” “should,” “could,” “would,” “may,” “will,” “forecast,” or the negative of these terms, and other similar expressions, although not all forward-looking statements contain these words. We base these forward-looking statements or projections, including Mobileye’s full-year guidance, on our current expectations, plans and assumptions that we have made in light of our experience in the industry, as well as our perceptions of historical trends, current conditions, expected future developments and other factors we believe are appropriate under the circumstances and at such time. You should understand that these statements are not guarantees of performance or results. The forward-looking statements and projections are subject to and involve risks, uncertainties and assumptions and you should not place undue reliance on these forward-looking statements or projections. Although we believe that these forward-looking statements and projections are based on reasonable assumptions at the time they are made, you should be aware that many factors could affect our actual financial results or results of operations and could cause actual results to differ materially from those expressed in the forward-looking statements and projections. Detailed information regarding these and other factors that could affect Mobileye’s business and results is included in Mobileye’s SEC filings, including the company’s Annual Report on Form 10-K for the year ended December 27, 2025, particularly in the section entitled “Item 1A. Risk Factors”. Copies of these filings may be obtained by visiting our Investor Relations website at \u003Ca href=\"http:\u002F\u002Fir.mobileye.com\" target=\"_blank\" rel=\"noopener noreferrer\">ir.mobileye.com\u003C\u002Fa> or the SEC’s website at \u003Ca href=\"http:\u002F\u002Fwww.sec.gov\" target=\"_blank\" rel=\"noopener noreferrer\">www.sec.gov\u003C\u002Fa>.\u003C\u002Fh6>\n\u003Cp> \u003C\u002Fp>","2026-06-16T04:00:00.000Z",1,"News, Amnon Shashua, Driverless MaaS, Industry, Autonomous Driving",{"id":38,"type":27,"lang":7,"primary_tag":39,"url":40,"title":41,"short_title":11,"subtitle":11,"description":42,"meta_description":42,"image":43,"img_alt":44,"author_name":11,"download_doc":11,"download_title":11,"download_btn":11,"webinar_video":11,"thumbnail":11,"author_id":11,"content":45,"publish_date":46,"featured":35,"is_gated":19,"is_infographic":11,"is_hidden":19,"is_active":20,"tags":47,"updated_at":23},321,"News","mobileye-secures-major-dms-production-program-with-leading-us-automaker","Mobileye斩获美国头部车企驾驶员监测系统（DMS）大型量产定点项目","Mobileye DMS™平台获全球车企青睐，此次新单将延续Mobileye车内感知业务增长势头","https:\u002F\u002Fstatic.mobileye.com\u002Fwebsite\u002Fus\u002Fcorporate\u002Fimages\u002F8bfc2d66bf2b488ba43e4ea74cf956e3_1774268681384.png","Mobileye DMS™能够联动驾驶员视线、注意力状态与实时道路情境，更智能地识别分心驾驶行为","\u003Cp>\u003Cspan data-contrast=\"auto\">\u003Cstrong>中国上海，2026年3月23日\u003C\u002Fstrong> — Mobileye（纳斯达克股票代码：MBLY）今日宣布，一家美国头部车企将在其搭载Mobileye EyeQ6L系统集成芯片的未来车型中集成Mobileye驾驶员监测系统（Mobileye DMS™），计划于2027年启动量产。这一新项目拓展了与该车企客户现有组合辅助驾驶系统（ADAS）项目的合作范围与功能配置，预计量产规模达数百万辆，将覆盖多款车型、多个年款。\u003C\u002Fspan>\u003C\u002Fp>\n\u003Cp>Mobileye车内感知平台涵盖驾驶员监测系统（DMS）与乘员监测系统（OMS），可与ADAS感知算法在单颗芯片上并行运行。平台能够融合车辆内部感知与车外道路感知，基于驾驶环境的具体情境来评估驾驶员参与度，不仅可判断驾驶员是否处于警觉状态，还能监测其视线方向，以及注意力是否与实时道路状况匹配。\u003C\u002Fp>\n\u003Cp>之前，Mobileye DMS和OMS已获一家全球车企青睐，将被集成到基于EyeQ6H的SuperVision™及环绕式ADAS™方案中，而此次新项目是此前业务的成功延续。这些项目共同体现出，车企日益希望将驾驶员监测、乘员安全与辅助驾驶功能整合，从而告别独立DMS电子控制单元（ECU）带来的额外成本与系统复杂度。\u003C\u002Fp>\n\u003Cp>Mobileye业务发展与战略执行副总裁Nimrod Nehushtan 表示：“新一代辅助驾驶需要车辆全维度、更丰富的情境信息，包括前方道路、车辆内部以及二者的交互关联。与此同时，车企希望在全系车型中规模化搭载辅助驾驶功能，同时避免额外的硬件成本与复杂的系统集成工作。Mobileye DMS兼顾两大需求，可在单一ADAS芯片及ECU平台上实现情境感知驾驶员监测。这样的功能集成正是Mobileye独有的优势，我们期待助力客户实现规模化量产落地。”\u003C\u002Fp>\n\u003Cp>随着运动脱离驾驶功能向非豪华车型普及，确保驾驶员切实关注道路状况，是该功能安全落地的关键。Mobileye DMS可联动ADAS摄像头采集的真实路况与驾驶员视线，识别仅靠车内独立系统无法捕捉的分心行为，同时判断驾驶员是否已感知路况。该系统能够减少误报，实现更精准的干预控制；针对更高等级的驾驶自动化，可基于驾驶员参与度，更智能地发出接管请求。\u003C\u002Fp>\n\u003Cp>该平台旨在满足欧洲新车安全评鉴协会（Euro NCAP）2026版评分要求，同时前瞻性地着眼于Euro NCAP 2029版规程的潜在升级方向，预计届时评测标准将从单纯的眼球追踪提升至有效驾驶参与度检测。\u003C\u002Fp>\n\u003Cp> \u003C\u002Fp>\n\u003Cp>\u003Cstrong>\u003Cspan data-contrast=\"auto\">About Mobileye\u003C\u002Fspan>\u003C\u002Fstrong>\u003Cspan data-ccp-props=\"{"201341983":0,"335559739":0,"335559740":240}\"> \u003C\u002Fspan>\u003C\u002Fp>\n\u003Cp>\u003Cspan data-contrast=\"auto\">Mobileye (Nasdaq: MBLY) leads the mobility revolution with our autonomous driving and driver-assistance technologies, harnessing world-renowned expertise in artificial intelligence, computer vision and integrated software and hardware. Since our founding in 1999, Mobileye has enabled the global adoption of advanced driver-assistance systems that save countless lives and reduce crashes, while pioneering groundbreaking technologies such as REM™ crowdsourced road intelligence, Imaging Radar and Compound AI. These technologies drive the ADAS and AV fields towards the future of mobility – enabling self-driving vehicles and mobility solutions at scale, and powering industry-leading ADAS products. Through 2025, about 230 million vehicles worldwide have been built with Mobileye’s EyeQ technology inside, and in 2026 Mobileye acquired Mentee Robotics to pursue the future of physical AI and humanoid robots. Since 2022, Mobileye has been listed independently from Intel (Nasdaq: INTC), which retains majority ownership. For more information, visit \u003C\u002Fspan>\u003Ca href=\"https:\u002F\u002Fcts.businesswire.com\u002Fct\u002FCT?id=smartlink&url=https%3A%2F%2Fwww.mobileye.com&esheet=54421863&newsitemid=20260210772101&lan=en-US&anchor=https%3A%2F%2Fwww.mobileye.com&index=1&md5=cfa04b95770b8044fedb93639c4fc56e\">\u003Cspan data-contrast=\"auto\">https:\u002F\u002Fwww.mobileye.com\u003C\u002Fspan>\u003C\u002Fa>\u003Cspan data-ccp-props=\"{"201341983":0,"335559739":0,"335559740":240}\"> \u003C\u002Fspan>\u003C\u002Fp>\n\u003Cp>\u003Cspan data-contrast=\"auto\">Mobileye's name, product names, product marks and logos are trademarks or registered trademarks of Mobileye Vision Technologies Limited. Other names and brands are the property of their respective owners.\u003C\u002Fspan>\u003Cspan data-ccp-props=\"{"201341983":0,"335559739":0,"335559740":240}\"> \u003C\u002Fspan>\u003C\u002Fp>\n\u003Cp>\u003Cspan data-contrast=\"auto\">Any reference to \"Mobileye\" in this document means Mobileye Vision Technologies Ltd., Mobileye Global Inc. or any of their subsidiaries.\u003C\u002Fspan>\u003Cspan data-ccp-props=\"{"201341983":0,"335559739":0,"335559740":240}\"> \u003C\u002Fspan>\u003C\u002Fp>","2026-03-23T05:00:00.000Z","ADAS",{"id":49,"type":6,"lang":7,"primary_tag":50,"url":51,"title":52,"short_title":11,"subtitle":11,"description":53,"meta_description":53,"image":54,"img_alt":55,"author_name":11,"download_doc":11,"download_title":11,"download_btn":11,"webinar_video":11,"thumbnail":11,"author_id":11,"content":56,"publish_date":57,"featured":35,"is_gated":19,"is_infographic":11,"is_hidden":19,"is_active":20,"tags":58,"updated_at":23},298,"Industry","where-mobility-meets-the-game-mobileye-partners-with-vfl-wolfsburg","智慧出行邂逅绿茵激情：Mobileye与德甲沃尔夫斯堡足球俱乐部达成战略合作","Mobileye 正式成为德国足球甲级联赛（Bundesliga）劲旅沃尔夫斯堡足球俱乐部的顶级合作伙伴。","https:\u002F\u002Fstatic.mobileye.com\u002Fwebsite\u002Fus\u002Fcorporate\u002Fimages\u002F3190dec398225b0d7744d1f29bcae4e6_1751278008967.png","沃尔夫斯堡足球俱乐部是一支独具特色的球队，始终秉持创新精神、前瞻视野与团队协作理念。","\u003Cp>打造卓越出行体验始终是Mobileye的使命核心。如今，我们首次将这份追求卓越的精神延伸至绿茵场，成为德甲劲旅沃尔夫斯堡俱乐部的顶级合作伙伴。\u003C\u002Fp>\n\u003Ch3>智慧出行与绿茵激情的完美碰撞\u003C\u002Fh3>\n\u003Cp>这项\u003Ca href=\"https:\u002F\u002Fwww.vfl-wolfsburg.de\u002Fen\u002Fnewsdetails\u002Fnews-detail\u002Fdetail\u002Fnews\u002Fautomotive-technology-meets-team-spirit-a-vision-for-the-future\">战略合作\u003C\u002Fa>对 Mobileye 意义重大，不仅关乎品牌本身，更体现了我们所坚持的理念。作为一家汽车科技公司，这次合作为我们在德国、欧洲乃至全球范围内进一步提升品牌影响力打开了新的机遇。\u003C\u002Fp>\n\u003Cp>沃尔夫斯堡足球队以创新、远见与团队精神著称，正是这些价值观，让这次的合作显得如此契合、自然。这不仅是两个领域的携手同行，更是一次共同塑造未来的旅程——发生在一个长期象征智慧出行与技术进步的国家与城市之中。\u003C\u002Fp>\n\u003Ch3>沃尔夫斯堡俱乐部与德甲联赛\u003C\u002Fh3>\n\u003Cp>沃尔夫斯堡足球俱乐部是德甲联赛的重要成员，而德甲是世界足坛最具影响力的联赛之一。该赛事不仅汇聚了众多国际知名球队，更涌现了无数闪耀足坛的传奇球星与当世巨星，因此，每逢周末，数以千万计的全球球迷都会将目光聚焦于德甲赛场。\u003C\u002Fp>\n\u003Cp>沃尔夫斯堡是德甲联赛的传统劲旅，曾在2009年问鼎德甲冠军，并于2015年捧起德国杯，是21世纪德国足坛最成功的俱乐部之一。俱乐部所在城市沃尔夫斯堡不仅是大众汽车集团总部所在地，更是德国汽车工业的重要象征。\u003C\u002Fp>\n\u003Ch3>品牌影响力跃升\u003C\u002Fh3>\n\u003Cp>随着 Mobileye 品牌正式走上绿茵赛场，我们将迎来前所未有的曝光机会。球衣品牌展示、场边广告横幅以及社交媒体传播，将全面提升 Mobileye 的品牌可见度。\u003C\u002Fp>\n\u003Cp>以下是球迷将在 2025\u002F2026 赛季看到 Mobileye 的一些重点展示场景：\u003C\u002Fp>\n\u003Cp>Mobileye 的logo将出现在沃尔夫斯堡队比赛球衣的左袖上，覆盖所有赛事，包括德甲、德国杯、国际赛事以及友谊赛。除了球衣展示，Mobileye 品牌还将在球场内外拥有强大的曝光版位，全面提升品牌可见度。\u003C\u002Fp>\n\u003Ch3>超越足球的战略意义\u003C\u002Fh3>\n\u003Cp>此次合作以建立深度联结为核心，是Mobileye深化产业布局的重要一步，展现了我们在各个领域追求卓越的坚定决心，旨在将Mobileye的创新故事传递给更多受众，共同见证科技创新与足球碰撞出的精彩火花。让我们期待新赛季的到来。\u003C\u002Fp>\n\u003Cp> \u003C\u002Fp>","2025-06-30T04:00:00.000Z","Events, Industry",{"links_array":60},[61,91,112,133,137],{"id":62,"name":63,"is_link":20,"is_local_link":20,"path":64,"sub_links":65,"is_icon":19},"1","公司信息","\u002Fabout\u002F",[66,68,72,75,79,83,87],{"id":62,"name":67,"is_local_link":20,"path":64},"关于我们",{"id":69,"name":70,"is_local_link":20,"path":71},"2","管理层","\u002Fabout\u002Fmanagement\u002F",{"id":73,"name":8,"is_local_link":20,"path":74},"3","\u002Famnon-shashua\u002F",{"id":76,"name":77,"is_local_link":19,"path":78},"4","投资人","https:\u002F\u002Fir.mobileye.com\u002F",{"id":80,"name":81,"is_local_link":20,"path":82},"5","ESG","\u002Fabout\u002Fesg\u002F",{"id":84,"name":85,"is_local_link":19,"path":86},"6","招贤纳士","https:\u002F\u002Fcareers.mobileye.com\u002F",{"id":88,"name":89,"is_local_link":20,"path":90},"7","联系我们","\u002Fcontact\u002F",{"id":69,"name":92,"is_link":20,"is_local_link":20,"path":93,"sub_links":94,"is_icon":19},"解决方案","\u002Fsolutions\u002F",[95,97,100,103,106,109],{"id":62,"name":96,"is_local_link":20,"path":93},"解决方案概览",{"id":69,"name":98,"is_local_link":20,"path":99},"Mobileye ADAS","\u002Fsolutions\u002Fadas\u002F",{"id":73,"name":101,"is_local_link":20,"path":102},"Mobileye SuperVision™","\u002Fsolutions\u002Fsuper-vision\u002F",{"id":76,"name":104,"is_local_link":20,"path":105},"Mobileye Chauffeur™","\u002Fsolutions\u002Fchauffeur\u002F",{"id":80,"name":107,"is_local_link":20,"path":108},"Mobileye Drive™","\u002Fsolutions\u002Fdrive\u002F",{"id":84,"name":110,"is_local_link":20,"path":111},"EyeQ Kit™","\u002Fsolutions\u002Feyeq-kit\u002F",{"id":73,"name":113,"is_link":20,"is_local_link":20,"path":114,"sub_links":115},"尖端技术","\u002Ftechnology\u002F",[116,118,121,124,127,130],{"id":62,"name":117,"is_local_link":20,"path":114},"技术概览",{"id":69,"name":119,"is_local_link":20,"path":120},"EyeQ™ 系统集成芯片","\u002Ftechnology\u002Feyeq-chip\u002F",{"id":73,"name":122,"is_local_link":20,"path":123},"真正冗余","\u002Ftechnology\u002Ftrue-redundancy\u002F",{"id":76,"name":125,"is_local_link":20,"path":126},"智能路网技术 (REM™)","\u002Ftechnology\u002Frem\u002F",{"id":80,"name":128,"is_local_link":20,"path":129},"责任敏感安全 (RSS)","\u002Ftechnology\u002Fresponsibility-sensitive-safety\u002F",{"id":84,"name":131,"is_local_link":20,"path":132},"安全方法论","\u002Ftechnology\u002Fsafety-methodology\u002F",{"id":76,"name":134,"is_link":20,"is_local_link":20,"path":135,"sub_links":136,"is_icon":19},"CEO专栏","\u002Fceo-corner\u002F",[],{"id":80,"name":138,"is_link":20,"is_local_link":20,"path":139,"sub_links":140},"新闻中心","\u002Fnews\u002F",[141,143],{"id":62,"name":142,"is_local_link":20,"path":139},"新闻",{"id":69,"name":144,"is_local_link":20,"path":145},"博客","\u002Fblog\u002F",{"linksList":147,"socialLinks":199},[148,158,165,172,174,178],{"id":62,"link":149,"name":63,"isLocalLink":20,"subLinks":150},"\u002F",[151,152,153,154,155,156,157],{"id":62,"link":64,"name":67,"isLocalLink":20},{"id":69,"link":71,"name":70,"isLocalLink":20},{"id":73,"link":74,"name":8,"isLocalLink":20},{"id":76,"link":78,"name":77,"isLocalLink":19},{"id":80,"link":82,"name":81,"isLocalLink":20},{"id":84,"link":86,"name":85,"isLocalLink":19},{"id":88,"link":90,"name":89,"isLocalLink":20},{"id":69,"link":93,"name":92,"isLocalLink":20,"subLinks":159},[160,161,162,163,164],{"id":62,"name":98,"isLocalLink":20,"link":99},{"id":69,"name":101,"isLocalLink":20,"link":102},{"id":73,"name":104,"isLocalLink":20,"link":105},{"id":76,"name":107,"isLocalLink":20,"link":108},{"id":80,"name":110,"isLocalLink":20,"link":111},{"id":73,"link":114,"name":113,"isLocalLink":20,"subLinks":166},[167,168,169,170,171],{"id":62,"name":119,"isLocalLink":20,"link":120},{"id":69,"name":122,"isLocalLink":20,"link":123},{"id":73,"name":125,"isLocalLink":20,"link":126},{"id":76,"name":128,"isLocalLink":20,"link":129},{"id":80,"name":131,"isLocalLink":20,"link":132},{"id":76,"link":135,"name":134,"isLocalLink":20,"subLinks":173},[],{"id":80,"link":149,"name":138,"isLocalLink":20,"subLinks":175},[176,177],{"id":62,"link":145,"name":144,"isLocalLink":20},{"id":69,"link":139,"name":142,"isLocalLink":20},{"id":84,"link":149,"name":179,"isLocalLink":20,"subLinks":180},"法务条款",[181,184,187,190,193,196],{"id":62,"link":182,"name":183,"isLocalLink":19},"https:\u002F\u002Fwww.mobileye.com\u002Fterms-of-use\u002F","使用条款",{"id":69,"link":185,"name":186,"isLocalLink":19},"https:\u002F\u002Fwww.mobileye.com\u002Fglobal-candidate-privacy-notice\u002F","候选人隐私声明",{"id":73,"link":188,"name":189,"isLocalLink":19},"https:\u002F\u002Fwww.mobileye.com\u002Fsecurity-and-compliance\u002F","安全与合规",{"id":76,"link":191,"name":192,"isLocalLink":19},"https:\u002F\u002Fwww.mobileye.com\u002Fprivacy-policy\u002F","隐私政策",{"id":80,"link":194,"name":195,"isLocalLink":19},"https:\u002F\u002Fwww.mobileye.com\u002Fcookies-notice\u002F","Cookies通知",{"id":84,"link":197,"name":198,"isLocalLink":19},"https:\u002F\u002Fbrand.mobileye.com\u002Fd\u002FaLZ6VeziuWM7\u002Flegal","Logo及标志使用",[200,204,208,211],{"id":62,"link":201,"icon":202,"isLocalLink":19,"withQr":20,"qrIcon":203},"https:\u002F\u002Fweixin.qq.com\u002Fr\u002FrTju9qbELPtTreKV922-","https:\u002F\u002Fstatic.mobileye.com\u002Fwebsite\u002Fcorporate\u002Fimg\u002Fhomepage2022\u002Fcn_wechat_icon.svg","https:\u002F\u002Fstatic.mobileye.com\u002Fwebsite\u002Fcn\u002Fcorporate\u002Ficons\u002Fmobileye_wechat_qr_small.png",{"id":69,"link":205,"icon":206,"isLocalLink":19,"withQr":19,"qrIcon":207},"https:\u002F\u002Fweibo.com\u002Fmobileye","https:\u002F\u002Fstatic.mobileye.com\u002Fwebsite\u002Fcorporate\u002Fimg\u002Fhomepage2022\u002Fcn_weibo_icon.svg","",{"id":73,"link":209,"icon":210,"isLocalLink":19,"withQr":19,"qrIcon":207},"https:\u002F\u002Fwww.zhihu.com\u002Forg\u002Fmobileye-wu-bi-shi","https:\u002F\u002Fstatic.mobileye.com\u002Fwebsite\u002Fcorporate\u002Fimg\u002Fhomepage2022\u002Fcn_zhihu_icon_2.svg",{"id":76,"link":212,"icon":213,"isLocalLink":19,"withQr":19,"qrIcon":207},"https:\u002F\u002Fspace.bilibili.com\u002F1730991723","https:\u002F\u002Fstatic.mobileye.com\u002Fwebsite\u002Fcorporate\u002Fimg\u002Fhomepage2022\u002Fcn_bilibili_icon_2.svg",2026,1785839814815]