The BYD “Han” model. [BYD]
While Level 2 autonomy (SAE-defined) requires driver supervision (For context, at Level 2, a vehicle can simultaneously control steering, acceleration, and braking in specific scenarios like highways or traffic jams, but the driver remains in control.), BYD’s DiPilot also weaves in AI-driven capabilities and scalable hardware with potential upgrades to follow. For instance, the entry-level DiPilot 100 supports highway navigation, lane changes, and obstacle avoidance. BYD plans on adding urban driving via over-the-air (OTA) updates. It also includes remote valet parking with 2 cm parking accuracy, even in the company’s $9,550 Seagull model, plus autonomous emergency braking (AEB) at 100 km/h (about 62 MPH).
For the sake of comparison, competitors like Tesla charge approximately $8,200 for its Full Self-Driving (FSD) package on pricier models.
The move pressures rivals like Tesla and China-based Xpeng. After announcing the news, Tesla stock dipped about
DeepSeek Integration
At the core of BYD’s DiPilot system is DeepSeek’s R1 model, a large-scale AI engine originally developed for language processing but adapted to handle the demands of real-time driving. Unlike traditional rule-based computer vision, R1 uses neural networks to interpret the environment, detecting and classifying objects—cars, pedestrians, lane markers—in varied weather or lighting conditions. Beyond perception, R1 can support decision-making and help predict possible maneuvers such as lane changes or braking. It can also spot road hazards and make split-second risk assessments.
By integrating DeepSeek’s R1 reasoning model, BYD gains AI-powered perception that handles complex urban scenarios – from traffic light recognition to chaotic intersections. The system draws on 72 million km of daily cloud training–based data and relies on centralized architecture to make split-second decisions. The DiPilot system is based on BYD’s Xuanji Architecture, which it describes as having “one brain, two ends, three networks, and four chains.” In other words, it integrates a central processor, cloud AI, vehicle-side AI, and sensor networks, all linked to DeepSeek R1 to support real-time decision-making capability.
DiPilot integrates AI-driven perception in three hardware tiers (including LiDAR on premium models — for more, read on), all receiving OTA updates. BYD’s strategic advantage lies in feeding the R1 model a continual stream of real-world driving data from every DiPilot-equipped vehicle. Cameras, radar, and sensors capture a range of driving situations while preserving driver anonymity. This large dataset fuels an iterative “flywheel” cycle: more vehicles generate more training data, refining the AI and attracting more buyers.
A breakdown of the three-tiered architecture
BYD currently offers its DiPilot ADAS in three main hardware tiers—DiPilot 100, 300, and 600:
- DiPilot 100 (Entry Tier): Relies on a camera-based approach supplemented by radar and 12 ultrasonic sensors. BYD has disclosed a total of 12 cameras, 5 millimeter-wave radars, and 12 ultrasonic units for this tier. Although it omits LiDAR, DiPilot 100 supports Level-2 features like highway NOA (Navigate-on-Autopilot), lane changes, and remote self-parking. BYD aims to expand city NOA to this tier via OTA updates.
- DiPilot 300 (Mid Tier): Adds a single LiDAR unit for more robust 3D mapping and object detection. This enables advanced city navigation, improved obstacle detection in poor visibility, and potentially more sophisticated parking maneuvers. The camera, radar, and ultrasonic setup is similar, but the LiDAR integration and software calibration allow for expanded autonomy features.
- DiPilot 600 (Premium Tier): Incorporates multiple LiDAR sensors (in some models, three LiDAR units) to provide panoramic coverage of the vehicle’s surroundings. Designed for flagship models like the Yangwang U9, DiPilot 600 can reportedly handle complex urban scenarios and high-speed highways with greater redundancy and 360° scanning.
Each tier is built upon BYD’s Xuanji Architecture, which centralizes all sensor data into a single processing “brain.”
Sources: BYD shares hit record high after EV maker rolls out driver assistance tech with DeepSeek’s AI help – NBC New York; BYD’s new self-driving tech aims to rival Tesla with DeepSeek AI integration; BYD’s Answer To Tesla Autopilot Is Way, Way Cheaper; Xpeng, Geely Shares Fall After BYD Launches Advanced Driver Assistance Features On Car Priced Below $10,000; China’s BYD to add DeepSeek AI to its affordable EVs | Latest Market News.
