特斯拉 FSD（輔助監督版）累計里程破 80 億英里

Post LinkedIn

🏠閱讀原文： IT之家

#autonomous-driving#fleet-data#long-tailtesla-fsd-supervised

💡Tesla FSD's 8B miles nears 10B unsupervised threshold—vital data scale benchmark for AV AI researchers.

⚡ 30-Second TL;DR

有什麼變化

車主駕駛 FSD 輔助監督版總里程超 80 億英里（約 128.75 億公里）。

為什麼重要

提升特斯拉在自動駕駛數據規模領先，迫使競爭對手擴大車隊數據收集。透過真實里程而非模擬，更好處理罕見邊緣案例。

下一步行動

Benchmark your AV dataset growth against Tesla's 1B-mile jump in months using fleet telemetry tools.

誰應關注：Researchers & Academics

關鍵要點

•車主駕駛 FSD 輔助監督版總里程超 80 億英里（約 128.75 億公里）。
•於 2024 年 12 月 27 日達 70 億英里後數月達成，增速加快。
•數據捕捉真實長尾場景，用於端到端 AI 模型訓練。
•馬斯克：100 億英里為安全無監督 FSD 門檻。

🧠 深度解析

背景與延伸：來自公開資料，非原文內容。引用 7 個來源。

🔑 增強重點摘要

•Tesla FSD Supervised has surpassed 8.2 billion miles driven by owners, with over 3 billion miles in city conditions, accelerating from 7 billion miles reached on December 27, 2025[1][2][3].
•Safety data shows FSD Supervised achieves 7X fewer major and minor collisions and 5X fewer off-highway collisions compared to U.S. national averages, with 5.3 million miles per crash in North America[1][2][5].
•The system processes over one million pixels of visual data every millisecond via eight external cameras for a 360-degree real-time view, improving road safety by over 80%[2].
•Elon Musk highlighted the milestone on X, noting substantial miles driven, while targeting 10 billion miles of real-world data to handle long-tail scenarios for unsupervised FSD[1][3].
•FSD is available via $99/month subscription, with outright purchase still offered for Model S/X, amid announcements of Cybercab production at $30,000 or less before 2027[2][3][6].

📊 競品分析▸ Show

Feature	Tesla FSD Supervised	Waymo
Miles Driven	8.2B+ (supervised, customer fleet)	Not specified in results; Tesla trails in some unsupervised benchmarks [5]
Miles per Major Collision	5.3M (North America)	U.S. avg: 0.66M [5]
Pricing	$99/mo subscription; outright for some models	Ride-hailing service, no ownership pricing [5]
Status	Supervised (SAE Level 2)	Unsupervised robotaxi in select areas

Note: Limited direct competitor benchmarks in results; Waymo mentioned as leading in unsupervised ops[5].

🛠️ 技術深入

•FSD Supervised uses end-to-end AI models trained on real-world fleet data capturing long-tail scenarios (rare, complex events) beyond simulations[3].
•Vision-based system with eight external cameras providing 360-degree coverage, processing >1 million pixels per millisecond for real-time environmental response[2].
•Passive and active safety features avoid ~1.9 million potential injuries annually, with >80% road safety improvement claimed[6].
•Cumulative data differentiates from total FSD miles; supervised miles heavily contribute to training for unsupervised capability[3].

🔮 前景展望AI analysis grounded in cited sources

The 8.2B mile milestone bolsters Tesla's data advantage for AI training, potentially enabling unsupervised FSD at 10B miles, accelerating robotaxi deployment like Cybercab and challenging competitors in scalable autonomy while highlighting supervised safety gains over human drivers.