China Autonomous Driving & Intelligent Connected Vehicle Technology Development: A Deep Dive Report

Date: April 12, 2026 Category: EV Technology & Innovation Depth: Deep Research Report Data Sources: 30+ content_library research documents, 28 evidence items, 10 observations, 5 cross-analyses

⚠️ Limited data coverage: No structured/tabular data (excel_assets) is available for this category. All findings are sourced from analyst research reports and industry tracking documents in the content_library index. Readers should treat numerical projections as research estimates rather than audited statistics.

I. Executive Summary

China's autonomous driving industry reached a decisive inflection point in late 2025. The Ministry of Industry and Information Technology (MIIT) granted the nation's first L3 conditional autonomous driving access permits, transitioning the technology from road testing to regulated commercial operation.[^1][^2] This regulatory breakthrough, combined with rapid advances in end-to-end AI architectures, falling hardware costs, and accelerating Robotaxi fleet deployments, marks 2026 as the year China's autonomous driving industry shifts from technology validation to规模化落地 (规模化 deployment).[^3]

Key findings:

1. L3 commercialization is real and measured. The approved Deepal L3 model completed over 70,000 km of autonomous driving in 19 days across complex urban scenarios.[^3] The liability framework — shifting responsibility from driver to automaker upon system activation — resolves the core ambiguity that held back L2 systems.[^4]

2. AI architecture is the primary capability driver. End-to-end neural network architectures (Huawei ADS 4.0, Li Auto Mind VLA, NVIDIA Alpamayo) have reduced latency by 50% and improved通行效率 by 20%, while open-source models are democratizing development access.[^3][^7]

3. Hardware cost curves are enabling "smart driving democratization." LiDAR has dropped from the 400K-500K yuan vehicle segment (2022) to 250K-300K (2024), with 1,000-yuan units targeting 100K-yuan vehicles.[^8] BYD's deployment of advanced ADAS on sub-100K-yuan vehicles signals mass-market access.[^8]

4. Robotaxi is at the profitability inflection. China's top three operators (Pony.ai, Apollo Go, WeRide) each operate 1,000+ vehicle fleets, with some cities achieving monthly per-vehicle profitability.[^9][^10][^11] The sector is projected to grow at 247% CAGR through 2030.[^18]

5. The value chain is structurally shifting upstream. Computing chips and domain controllers now capture 25-30% of vehicle value, while the traditional OEM assembly margin shrinks relative to software and data services.[^22]

II. Policy & Regulatory Framework: From Testing to Commercialization

2.1 L3 Access Permits: The Regulatory Breakthrough

On December 15, 2025, MIIT formally announced China's first L3 conditional autonomous driving vehicle access permits, approving two models: the BAIC Arcfox Alpha S5 for Beijing highway operation (maximum 80 km/h) and the Changan Deepal SL03 for Chongqing operation (maximum 50 km/h).[^1][^2] This was not another testing permit — it was a product access permit, signifying the transition from R&D validation to regulated commercial delivery.[^2]

The regulatory approach reflects a deliberate "start small, expand gradually" philosophy. Both approved models are restricted to designated highway sections and rapid transit roads, with speed caps calibrated to manage kinetic energy in failure scenarios.[^2] At lower speeds, braking distances are shorter, accident severity is reduced, and the pressure on redundant systems (braking, steering, power supply) drops exponentially.[^2]

The pilot performance data validates this cautious approach. Deepal's L3 vehicles accumulated over 70,000 km of autonomous driving in just 19 days, covering complex urban scenarios including elevated interchanges and congested road segments.[^3] Changan's validation program covered over 400,000 simulation scenarios, 191 types of field test scenarios (10x the national standard), and over 5 million km of road实证 with 36% classified as extreme scenarios.[^5]

2.2 Liability Framework: Resolving the Core Ambiguity

The most consequential element of the L3 framework is the liability shift. Under China's《汽车驾驶自动化分级》, L0-L2 systems are classified as driving assistance, with the driver as the sole liable party. L3 represents the first level at which the automaker or system supplier becomes the responsible entity when the system is activated.[^4]

This change addresses the fundamental "responsibility ambiguity" problem that plagued L2 systems, where unclear liability discouraged both full system utilization and insurance product development.[^4] The new framework requires automakers to prove system non-defect in the event of an accident, inverting the burden of proof.[^4]

The pilot strategy follows a "B-end first, gradually open to individual users" approach. Approved L3 models are currently limited to pilot operating units and are not available for individual purchase.[^2] L3 models are expected to begin rolling out to individual users in Q2 2026.[^3]

2.3 Mandatory National Standards for L3/L4

In February 2026, MIIT published draft mandatory national standards for L3/L4 autonomous driving system safety requirements, opening them for public comment.[^13] The standards establish that autonomous driving systems must perform at least at the level of a qualified, attentive human driver, with specific requirements for:

• Nominal scenario collision avoidance with safety-relevant targets[^13]

• Safe and effective interaction with other road users (turn signals, brake lights)[^13]

• Compliance with traffic regulations without unreasonably disrupting traffic flow[^13]

• Detection and response to priority vehicles (police, fire, ambulance)[^13]

• Response to on-site traffic police direction[^13]

These standards reference UN Regulation No. 157 (ALKS) and are expected to take effect on July 1, 2027.[^13] Additionally, the《自动驾驶汽车运营服务规范》(Autonomous Driving Vehicle Operation Service Standards), initiated in July 2024 with an 18-month project cycle, is expected to finalize in 2026, establishing requirements for vehicle management, personnel staffing, safety assurance, and service delivery.[^14]

2.4 Local City-Level Regulations: A Patchwork of Innovation

Five major Chinese cities have enacted comprehensive ICV (Intelligent Connected Vehicle) regulations, each with distinct approaches:

• Shenzhen (Aug 2022): First to define L3-L5 classifications, permit fully driverless vehicles in designated areas, and establish a "driver/safety officer liability determination" mechanism. Effects: Pony.ai and Yuanrong Qixing received permits for revenue-generating operations; Mawan Port achieved commercial autonomous container truck operations in 2025.[^26]

• Guangzhou (Feb 2025): Permits operations across urban/highway/airport scenarios, mandates "vehicle-road-cloud integration" construction, and established mixed-traffic pilot zones. Effects: 3,240 km² operational area opened; WeRide launched nighttime autonomous BRT dedicated lines; Pony.ai opened intercity routes between Baiyun Airport and Guangzhou South Station.[^26]

• Wuhan (Jun 2022, upgraded Mar 2025): Enabled fully driverless testing and收费 operations with a "black box" system recording 90 seconds pre-accident data. Effects: Apollo Go deployed 400+ vehicles in 2024 with 10K+ daily orders; achieved收支平衡 (breakeven) by March 2025. The upgraded regulation mandates municipal platform data management, establishes vehicle owner liability for unstaffed accidents, and requires liability insurance of no less than 5 million yuan.[^26]

• Beijing (Apr 2025): Supports taxi, bus, and personal vehicle scenarios, requires infrastructure for "vehicle-road coordination," and establishes a Beijing-Tianjin-Hebei policy mutual recognition mechanism. Effects: Yizhuang Economic Development Zone opened fully driverless operations; Apollo Go and Pony.ai cumulative testing exceeded 20 million km; pushing toward city-wide thousand-vehicle deployments.[^26]

• Shanghai (Jul 2025): Set a 2027 target of 6M+ L4 passenger trips, issued "demonstration operation" licenses permitting收费 operations, and pushed for Pudong full-district opening in 2025. Effects: 8 companies received first-batch licenses (including Baidu, Pony.ai); Pudong opened 1,000 km of test roads connecting Disney, airports, and core districts.[^26]

2.5 International Comparison

China's regulatory approach contrasts sharply with other major markets:

• Germany: The 2021 Autonomous Driving Act established clear liability boundaries (automaker responsible during L3 activation) and mandated data recording devices. Mercedes' DRIVE PILOT expanded speed range from 60 km/h to 95 km/h, permitting hands-free entertainment use.[^2]

• United States: Policy and capital resources are directed toward L4 Robotaxi, with L3 not becoming a mainstream transition path. Waymo operates fully driverless services in San Francisco, Phoenix, and Austin. The SELF DRIVE Act of 2026 proposes increasing annual exemptions for steering-wheel-less AVs from 2,500 to 90,000 vehicles, resolving state-level legislative fragmentation.[^14]

• Japan: Honda's Legend L3 model was discontinued due to high costs, limited usage scenarios, and low user adoption rates, reflecting a conservative commercial approach.[^2]

III. Technology Evolution: End-to-End AI and Open-Source Ecosystem

3.1 End-to-End Architecture: The Paradigm Shift

The autonomous driving industry has converged on end-to-end neural network architectures as the dominant technical path. Unlike traditional modular pipelines (perception → prediction → planning → control), end-to-end systems input raw sensor data directly and output steering angle, throttle, and brake commands through unified neural networks.[^6]

Huawei ADS 4.0 WEWA Architecture exemplifies this shift. Its two-component design includes:

1. Cloud World Engine: Generates over 100,000 virtual edge-case scenarios daily with corner-case density 1,000x that of the real world. This addresses the industry's long-tail data bottleneck by training decision models on synthetic scenarios (e.g., nighttime强光 interference, unexpected construction zone obstacles) rather than relying solely on limited human driving data.[^20]

2. Vehicle-Side World Behavior Model: Uses a Mixture of Experts (MoE) architecture to integrate multi-modal sensor data (LiDAR, cameras, mmWave radar) and directly output vehicle control commands, skipping the language conversion step. This achieves "perception-to-action" end-to-end mapping, reducing latency by 50%, improving通行效率 by 20%, and reducing hard braking rates by 30%.[^6][^20]

3.2 Open-Source Models: Democratizing Development

Two major open-source initiatives in early 2026 signal a structural shift in industry accessibility:

• NVIDIA Alpamayo: Released at CES 2026, this is the world's first open-source VLA (Vision-Language-Action) model with reasoning capability for autonomous driving. NVIDIA simultaneously open-sourced the AlpaSim high-fidelity simulation framework and a large-scale driving dataset, building a "model-simulation-data" open ecosystem.[^7]

• Horizon Robotics HoloBrain-0: Announced full open-source of its base model and framework in February 2026, providing Chinese automakers with an accessible alternative to proprietary solutions.[^28]

These open-source moves lower the R&D threshold for autonomous driving, particularly benefiting mid-tier automakers and technology startups that previously lacked access to training infrastructure and large-scale datasets.

3.3 Computing Power Requirements by Autonomy Level

The computing power required for autonomous driving scales dramatically with capability level:

This escalation drives demand for specialized automotive-grade SoCs and creates a strategic bottleneck around chip supply, thermal management, and power consumption optimization.[^21]

3.4 E/E Architecture Evolution

The vehicle's electronic/electrical (E/E) architecture is undergoing a three-stage evolution:

1. Distributed (Traditional): Each function has an independent ECU, with 100+ ECUs in premium vehicles and wiring harnesses exceeding 5 km. Software-hardware coupling makes OTA upgrades difficult.[^23]

2. Domain-Concentrated (Current): ECUs are integrated by functional domain (driving domain, cabin domain, body domain). Driving domain controllers process camera and radar data while running autonomous driving algorithms.[^23]

3. Central-Concentrated (Future): One or two high-compute central computing platforms (HPC) serve as the core, paired with Zone Control Units (ZCUs) for localized access. This enables hardware resource pooling and full software decoupling, supporting the software-defined vehicle paradigm.[^23]

Li Auto's open-source Star Ring OS (星环OS) exemplifies this transition, comprising four subsystems: intelligent driving system (brain control), intelligent vehicle system (cerebellum control), communication middleware (nervous system), and information security system (immune system).[^28]

IV. Industry Penetration: From L2 Mainstream to L3/L4 Emergence

4.1 L2 ADAS: Already Mainstream

As of January-July 2025, China's L2 ADAS-equipped passenger vehicle sales reached 7.76 million units, achieving a 62.6% penetration rate.[^15] Highway NOA assembly reached 1.102 million units in 25Q2 alone, growing 250% YoY with a 19.6% penetration rate, entering the mass deployment phase.[^15]

This widespread L2 adoption creates the data collection and user education foundation necessary for L3+ rollout.

4.2 Market Segmentation and Smart Feature Distribution

China's NEV market reached 10.15 million units in January-October 2025, with distinct smart feature distribution across price segments:

• Under 100K yuan (34.9% share): Cost-sensitive; smart features are basic.[^16]

• 100K-200K yuan (36.2% share): The highest growth potential for advanced smart features.[^16]

• Over 200K yuan (28.9% share): Mainstream new force models have saturated coverage with high-level smart features.[^16]

The 100K-200K yuan segment represents the critical battleground for "smart driving democratization."

4.3 BYD's "Tianshen Zhiyan" Strategy

In February 2025, BYD announced that all vehicle lines would carry the "Tianshen Zhiyan" (天神之眼) advanced intelligent driving system, explicitly deploying this technology to vehicles under 100,000 yuan.[^8] This move was enabled by:

• Mass production of LiDAR, high-pixel automotive CIS, high-compute driving chips, and One-Box wire-controlled braking systems

• Supply chain maturation driving down the total BOM cost for advanced ADAS

• AI large model and multi-sensor fusion technology iteration

The strategy signals the beginning of "smart driving equity" (智驾平权) — making advanced autonomous features accessible across all price segments.[^8]

4.4 L3 Pilot Performance and Validation

Changan Automobile's L3 validation program demonstrates the rigor required for commercial deployment:

• Simulation testing: 400,000+ scenarios covered[^5]

• Field testing: 191 scenario types, 10x the national standard[^5]

• Road实证: 5M+ km, with 36% extreme scenarios[^5]

• Cybersecurity: 182 network and data security专项 tests[^5]

• Functional safety: 21,000+ hours of expected functional safety with zero accidents[^5]

• Scenario library: 300+ critical scenarios for data-driven safety closed-loop optimization[^5]

4.5 2030 Forecast

Industry projections indicate:

• China smart car sales to exceed 30 million units by 2030[^17]

• L2-L3 penetration to approach 70% by 2026 before growth decelerates[^17]

• L4-L5 to accelerate penetration in 2026-2027, crossing the 5% threshold in 2027-2028[^17]

V. Robotaxi Commercialization: Fleet Scale, Economics, and Competition

5.1 China's Top Three Operators

Pony.ai (小马智行):

• Fleet: 1,159 vehicles by end of 2025; 2026 target of 3,000+ vehicles[^9]

• Milestone: Achieved monthly per-vehicle profitability in Guangzhou and Shenzhen[^9]

• Technology: 7th-generation system with 60-70% cost reduction; world's first automotive-grade solid-state LiDAR solution[^26]

• Cumulative road testing: 40M+ km[^26]

• Partnership: Joint venture with Toyota (丰准智能科技); Uber integration[^26]

Baidu Apollo Go (萝卜快跑):

• Service volume: 3.4M Robotaxi services globally in 25Q4 (+209% YoY); weekly order peak exceeding 300,000[^11]

• Cumulative orders: 11M+ globally[^26]

• Coverage: 15 cities including Beijing, Wuhan, Shanghai; expanding to Dubai and Abu Dhabi[^26]

• Latest: Received Hong Kong testing permit; planned 2026 deployment in Germany and UK at scale of thousands of vehicles[^26]

• Partnership: Volkswagen Traffic (Shanghai cooperation); Uber[^26]

WeRide (文远知行):

• Fleet: 1,023 vehicles by January 2026; 2026 global target of 2,600+ vehicles[^10]

• Revenue: Robotaxi business revenue of 45.9M yuan in 25Q2 (+837% YoY)[^26]

• Milestone: Expected per-vehicle breakeven in Abu Dhabi in 26Q1[^10]

• International: Largest Robotaxi fleet in the Middle East; 1.00 billion yuan追加 investment from Uber; targeting 15 additional international cities[^26]

• Partnership: Uber, Chery Automobile, Jinjiang Taxi[^26]

5.2 Global Benchmarks

Waymo:

• Fleet: 3,000+ vehicles by February 2026; planned expansion to 5,000-6,000 by year-end[^12]

• Orders: 400K weekly paid rides; targeting 1M+ weekly by year-end[^12]

• Cities: Operating in San Francisco, Los Angeles, Phoenix, Austin, Atlanta, and Miami[^12]

• Expansion: Added Dallas, Houston, San Antonio, and Orlando in February 2026, reaching 10 metro areas[^12]

• DAU/MAU: 203K DAU and 1.44M MAU as of February 2026 (+70% MoM)[^12]

• Funding: Completed $16B financing round at $126B valuation in February 2026[^12]

Tesla:

• Fleet: 447 tracked vehicles as of March 2026 (358 in Bay Area, 89 in Austin)[^12]

• Hardware: First steering-wheel-less, pedal-less CyberCab rolled off Texas Gigafactory in February 2026[^27]

• FSD Mileage: 8.3B cumulative miles of supervised FSD operation[^27]

• Expansion: Received permits in Arizona, Nevada, and New Hampshire; targeting Saudi Arabia deployment[^27]

5.3 Cost Economics

Robotaxi operational costs currently exceed those of human-driven ride-hailing due to expensive hardware (LiDAR), software, safety officers, and system redundancy. However, the per-kilometer cost is projected to reach parity with ride-hailing by 2026, after which Robotaxi costs are expected to decline further.[^18]

The Robotaxi market transaction value is projected to grow at a 247% CAGR from 2024 to 2030.[^18] By 2030, China's Robotaxi vehicle sales market is expected to exceed 230 billion yuan, with vehicle maintenance market exceeding 25 billion yuan.[^18]

5.4 Four Competitive Models

1. Tech Company-Led (Baidu, Pony.ai, WeRide): Technology iteration-driven scaling; competitive focus on cost reduction and internationalization (Middle East trials, automotive-grade hardware mass production).[^26]

2. Mobility Platform-Led (Didi, Caocao, Ruqi): Ecosystem integration building operational moats; competition centers on fleet network scale effects and mixed-dispatch models.[^26]

3. OEM-Led (XPeng, GAC, SAIC): Front-line mass production capability, customized vehicle R&D, open technology interfaces.[^26]

4. Traditional Enterprise Transformation (Taxi operators): License resources, operational experience, local market knowledge as comparative advantages.[^26]

OEMs with both mature advanced driving solutions and front-line mass production capabilities possess a structural advantage in transitioning to Robotaxi operations.[^26]

VI. Value Chain Restructuring: Computing Power, Domain Controllers, and Software-Defined Vehicles

6.1 Value Chain Distribution

The autonomous driving value chain is undergoing a structural shift. The decision layer (computing chips + domain controllers) now accounts for 25-30% of total vehicle value, representing the highest-margin segment of the supply chain.[^22] Meanwhile, the整车/ecosystem segment (OEM and services) retains 30-40% of value, with automakers controlling user access points and possessing data closed-loop, brand premium, and OTA monetization capabilities.[^22]

This mirrors the PC/mobile internet value curve, where chip manufacturers, technology providers (algorithms + data), and user operators occupy the two ends of the "smile curve," while traditional hardware producers occupy the lower-margin middle.[^8]

6.2 Computing Chip Market

Computing chip requirements escalate sharply by autonomy level (see Section 3.3). This creates strategic opportunities for:

• Horizon Robotics (地平线): Partnership with CATL subsidiary Times Intelligence for hardware-software integration; former CATL executive Zhu Wei appointed as Horizon president.[^12]

• Black Sesame (黑芝麻): First Huashan A2000 chip mass production project with Guoqi Zhikong, covering L2+ to L3 full-scenario intelligent driving, with first vehicles expected to mass produce in 2026.[^28]

6.3 Domain Controllers and Cabin-Driving Integration

Domain controllers are the key enabler of software-defined vehicles, with the driving domain and cabin domain being the core integration points. The trend toward跨域融合 (cross-domain integration) — combining driving and cabin functions into unified "舱驾一体" systems — is accelerating.[^23]

Major players include Desay SV, Joyson Electronics, KoboDa, Huayang Group, and Jingwei Hirain.[^12]

6.4 Huawei's Three-Tier Ecosystem

Huawei's Qiankun (乾崑) intelligent driving platform operates through three partnership models:

1. Smart Selection (HarmonyOS Intelligent Row): Huawei leads product definition, R&D, and channel sales. Partners: Seres (AITO M5/M7/M8/M9), Chery (Luxeed R7/R9), JAC (Maextro S800).[^20]

2. HI Mode: Full-stack solution including ADS intelligent driving, HarmonyOS, and three-electric systems. Partners: Avatr, Arcfox.[^20]

3. Standardized Components: Tier1 supplier model offering modular products (LiDAR, DriveOne electric drive).[^20]

This ecosystem approach enables Huawei to simultaneously compete at the premium end (Maextro at 700K-1M yuan) and the mid-market (Luxeed at 250K-330K yuan), while generating component revenue from non-partner OEMs.[^20]

6.5 LiDAR Cost Curve

LiDAR, the core sensor for city NOA-level intelligent driving, has seen dramatic cost reduction:

• 2022: Mainly deployed on vehicles in the 400K-500K yuan price range[^24]

• 2024: Penetrated the 250K-300K yuan segment through system-level cost reduction[^24]

• Future: 1,000-yuan LiDAR products expected to penetrate 100K-yuan vehicles, opening mass-market adoption[^24]

VII. Key Findings, Risks, and Future Outlook

7.1 Key Findings

1. 2026 is the convergence year. Technology maturity (end-to-end AI), regulatory clarity (L3 permits + mandatory standards), user acceptance (L2 at 62.6%), and business model validation (Robotaxi profitability) are all reaching critical mass simultaneously.[^3][^13][^15][^9]

2. The competition is no longer about individual features but ecosystem capability. Companies that combine algorithm self-development, chip design, data closed-loop, and front-line mass production — such as XPeng, Huawei's partner ecosystem, and leading Robotaxi operators — will capture disproportionate value.[^20][^26]

3. The value chain is permanently shifting from hardware to software and data. Domain controllers, computing chips, and OTA services represent the highest-growth, highest-margin segments, while traditional hardware assembly faces margin compression.[^22][^23]

4. China's regulatory approach is strategically differentiated from the US. While the US leapfrogs directly to L4, China is methodically building L3 capability with clear liability frameworks, creating a more structured path to full autonomy.[^2][^14]

7.2 Key Risks

• Technology delay risk: L3/L4 deployment progress falling short of expectations due to unresolved long-tail scenarios or sensor limitations.[^4]

• Regulatory uncertainty: National standards finalization timeline (expected 2027) may slip; local regulation fragmentation could impede cross-city operations.[^4]

• Competitive overcapacity: Industry "内卷" (involution) — excessive competition driving below-cost pricing — could undermine R&D investment sustainability across the supply chain.[^28]

• Core component cost risk: Slower-than-expected cost reduction in LiDAR, automotive-grade chips, or domain controllers could delay mass-market adoption.[^3]

• Supply chain disruption: Geopolitical tensions affecting semiconductor supply could constrain computing chip availability.[^14]

7.3 Future Outlook

The trajectory from AI-powered vehicles to autonomous mobility infrastructure is clear. By 2030, China's smart car market is projected to exceed 30 million units, with Robotaxi services representing over 30% of smart mobility transactions.[^17][^19] The industry will transition from "hardware-defined" to "software-defined" to "data-defined," with companies that build the most effective data closed-loops achieving sustainable competitive advantage.[^23]

The next 12-18 months will determine which players capture the infrastructure layer and which remain component suppliers. The convergence of technology readiness, regulatory clarity, and economic viability in 2026 makes this a decisive window.

Source References

[^1]: 爱建证券_L3自动驾驶准入报告 | File: 爱建证券_L3自动驾驶准入报告.pdf | Index: content_library | Doc ID: 2423d34f54acf31f7252a6ffa4225d299f9cacd1d2a89bf6753ceec547a3427a

[^2]: 爱建证券_L3自动驾驶准入报告 | File: 爱建证券_L3自动驾驶准入报告.pdf | Index: content_library | Doc ID: 2423d34f54acf31f7252a6ffa4225d299f9cacd1d2a89bf6753ceec547a3427a

[^3]: 金元证券_自动驾驶加速发展报告 | File: 金元证券_自动驾驶加速发展报告.pdf | Index: content_library | Doc ID: d8404d004b6e6a1c4e3e345c2144909611d04c3f850fb64dad9d1ba4b9397334

[^4]: 爱建证券_L3自动驾驶准入报告 | File: 爱建证券_L3自动驾驶准入报告.pdf | Index: content_library | Doc ID: 2423d34f54acf31f7252a6ffa4225d299f9cacd1d2a89bf6753ceec547a3427a

[^5]: 爱建证券_L3自动驾驶准入报告 | File: 爱建证券_L3自动驾驶准入报告.pdf | Index: content_library | Doc ID: 2423d34f54acf31f7252a6ffa4225d299f9cacd1d2a89bf6753ceec547a3427a

[^6]: 金元证券_自动驾驶加速发展报告 | File: 金元证券_自动驾驶加速发展报告.pdf | Index: content_library | Doc ID: d8404d004b6e6a1c4e3e345c2144909611d04c3f850fb64dad9d1ba4b9397334

[^7]: 金元证券_自动驾驶加速发展报告 | File: 金元证券_自动驾驶加速发展报告.pdf | Index: content_library | Doc ID: d8404d004b6e6a1c4e3e345c2144909611d04c3f850fb64dad9d1ba4b9397334

[^8]: 金元证券_自动驾驶加速发展报告 | File: 金元证券_自动驾驶加速发展报告.pdf | Index: content_library | Doc ID: d8404d004b6e6a1c4e3e345c2144909611d04c3f850fb64dad9d1ba4b9397334

[^9]: robotaxi_tracking_huayuan | File: robotaxi_tracking_huayuan.pdf | Index: content_library | Doc ID: b99c2524d6febeaf44b8adf3cb435402e4cbce15d8cf06498d547cde1347ef93

[^10]: robotaxi_tracking_huayuan | File: robotaxi_tracking_huayuan.pdf | Index: content_library | Doc ID: b99c2524d6febeaf44b8adf3cb435402e4cbce15d8cf06498d547cde1347ef93

[^11]: robotaxi_tracking_huayuan | File: robotaxi_tracking_huayuan.pdf | Index: content_library | Doc ID: b99c2524d6febeaf44b8adf3cb435402e4cbce15d8cf06498d547cde1347ef93

[^12]: ai_auto_mar_dwzq | File: ai_auto_mar_dwzq.pdf | Index: content_library | Doc ID: 129ad2760a7d9883ee45305ab308d823ae2447e7f1b8b2d6fad166642d8162d7

[^13]: robotaxi_tracking_huayuan | File: robotaxi_tracking_huayuan.pdf | Index: content_library | Doc ID: 2ac6eaa1bdc3676030afff396bebb5351d9fa6d954e0cf04e1745c4809c589f5

[^14]: robotaxi_tracking_huayuan | File: robotaxi_tracking_huayuan.pdf | Index: content_library | Doc ID: 2ac6eaa1bdc3676030afff396bebb5351d9fa6d954e0cf04e1745c4809c589f5

[^15]: 爱建证券_L3自动驾驶准入报告 | File: 爱建证券_L3自动驾驶准入报告.pdf | Index: content_library | Doc ID: 2423d34f54acf31f7252a6ffa4225d299f9cacd1d2a89bf6753ceec547a3427a

[^16]: 爱建证券_L3车型准入智能汽车加速 | File: 爱建证券_L3车型准入智能汽车加速.pdf | Index: content_library | Doc ID: 60b59d462b83a9f6fa4447fbd4e80e86fd79da0aebd3f5162211dca85e4b650b

[^17]: 爱建证券_L3车型准入智能汽车加速 | File: 爱建证券_L3车型准入智能汽车加速.pdf | Index: content_library | Doc ID: e120ac253b53fd126ae027b9a7367e2c0a5c9db2f707440d64cff9044fe236e8

[^18]: 爱建证券_L3车型准入智能汽车加速 | File: 爱建证券_L3车型准入智能汽车加速.pdf | Index: content_library | Doc ID: e120ac253b53fd126ae027b9a7367e2c0a5c9db2f707440d64cff9044fe236e8

[^19]: 爱建证券_L3车型准入智能汽车加速 | File: 爱建证券_L3车型准入智能汽车加速.pdf | Index: content_library | Doc ID: 048914da2dd1f946e89eb2c1a0f5cd7b4ec933981e27041d2643bbf6e2990525

[^20]: 爱建证券_L3车型准入智能汽车加速 | File: 爱建证券_L3车型准入智能汽车加速.pdf | Index: content_library | Doc ID: 0c92fb77c1f185562608f13179bf036d23c7ad155b83311777887f260221163c

[^21]: 爱建证券_L3自动驾驶准入报告 | File: 爱建证券_L3自动驾驶准入报告.pdf | Index: content_library | Doc ID: 2423d34f54acf31f7252a6ffa4225d299f9cacd1d2a89bf6753ceec547a3427a

[^22]: 金元证券_自动驾驶加速发展报告 | File: 金元证券_自动驾驶加速发展报告.pdf | Index: content_library | Doc ID: d8404d004b6e6a1c4e3e345c2144909611d04c3f850fb64dad9d1ba4b9397334

[^23]: 爱建证券_L3车型准入智能汽车加速 | File: 爱建证券_L3车型准入智能汽车加速.pdf | Index: content_library | Doc ID: f6df25beec6a364f0a1b00cfd378136134a485087ce7048020776d37bb29a2e7

[^24]: 爱建证券_L3自动驾驶准入报告 | File: 爱建证券_L3自动驾驶准入报告.pdf | Index: content_library | Doc ID: 2423d34f54acf31f7252a6ffa4225d299f9cacd1d2a89bf6753ceec547a3427a

[^25]: 爱建证券_L3车型准入智能汽车加速 | File: 爱建证券_L3车型准入智能汽车加速.pdf | Index: content_library | Doc ID: 048914da2dd1f946e89eb2c1a0f5cd7b4ec933981e27041d2643bbf6e2990525

[^26]: 爱建证券_L3车型准入智能汽车加速 | File: 爱建证券_L3车型准入智能汽车加速.pdf | Index: content_library | Doc ID: e120ac253b53fd126ae027b9a7367e2c0a5c9db2f707440d64cff9044fe236e8

[^27]: ai_auto_mar_dwzq | File: ai_auto_mar_dwzq.pdf | Index: content_library | Doc ID: 8b9352be0d56c00d13dd443042fe98bf17babb8ae36817ffd47840263a0aaf19

[^28]: ai_auto_mar_dwzq | File: ai_auto_mar_dwzq.pdf | Index: content_library | Doc ID: b5beccc93e94af4b8aa017afbdde83b228b91e4881a660aa741b07a4e7d15c67

Data Inventory Summary

Weak data coverage dimensions:

• No structured/tabular data for hardware cost tracking, fleet utilization metrics, or financial performance data

• Robotaxi profitability claims are based on operator statements and analyst estimates, not independently audited financials

• Chip computing requirements are industry estimates; specific product roadmaps from chip manufacturers would strengthen analysis

• LiDAR cost figures are directional; detailed BOM breakdowns are not available in current data sources