滴滴達成 L4 自主可控自動駕駛技術

🔑 增強重點摘要

• •DiDi's R2 Robotaxi utilizes a proprietary 'Beiyao' (北曜) Beta sensor suite, which integrates high-resolution LiDAR and 4D imaging radar to reduce hardware costs by approximately 60% compared to previous generations.
• •The company has shifted its strategy toward a 'hardware-software decoupling' model, allowing the autonomous driving stack to be integrated into various OEM platforms beyond GAC Aion, facilitating faster fleet scaling.
• •DiDi has established a dedicated 'Safety-First' data loop that utilizes edge computing to process complex urban traffic scenarios in real-time, specifically targeting the 'long-tail' edge cases that previously hindered L4 scalability.

🛠️ 技術深入

• •Architecture: Employs a transformer-based perception model that fuses multi-modal sensor data (LiDAR, Radar, Camera) at the feature level rather than the object level.
• •Compute Platform: Utilizes a custom-designed domain controller optimized for low-latency inference, supporting over 500 TOPS of processing power.
• •Safety System: Features a triple-redundant braking and steering architecture, ensuring the vehicle can reach a 'minimal risk condition' (MRC) even in the event of total primary system failure.
• •Mapping: Implements a 'light-map' approach that relies on real-time semantic localization, reducing the dependency on high-definition (HD) map updates for every road change.

🔮 前景展望基於引用來源的 AI 分析

⏳ 時間線