BYD and DeepSeek Introduce Affordable Autonomous Driving Features

The Chinese electric vehicle (EV) manufacturer BYD is transforming the automotive landscape by making advanced driver-assistance systems (ADAS) more affordable. The company has unveiled its "God's Eye" technology, also referred to as the DiPilot system, which offers Level-2 autonomous driving features as standard equipment on electric vehicles priced below $10,000.

Level 2 autonomy, as defined by the Society of Automotive Engineers (SAE), allows a vehicle to control steering, acceleration, and braking in specific conditions such as highways and stop-and-go traffic. However, the driver must remain attentive and ready to take control. The DiPilot system incorporates AI-driven functionalities and hardware that can be upgraded in the future. For example, the entry-level DiPilot 100 provides support for highway navigation, lane changes, and obstacle avoidance. BYD plans to enhance this with urban driving capabilities through over-the-air (OTA) updates. The system also includes remote valet parking with a parking accuracy of 2 cm, even in models like the $9,550 Seagull, and features autonomous emergency braking (AEB) that works at speeds of up to 100 km/h (about 62 MPH).

In comparison, competitors such as Tesla offer their Full Self-Driving (FSD) package at around $8,200 but only for higher-priced models, which puts BYD's pricing strategy in a competitive light.

DeepSeek Technology Integration

The heart of BYD's DiPilot system is powered by DeepSeek's R1 model. Initially developed for language processing, this large-scale AI engine has been adapted for real-time driving applications. Unlike traditional computer vision systems that rely on preset rules, the R1 uses neural networks to interpret various driving scenarios. This includes detecting and classifying objects like vehicles, pedestrians, and road markings in diverse weather or lighting conditions. In addition to perception, the R1 can assist in decision-making processes and help anticipate necessary maneuvers such as lane changes or stopping for hazards.

By integrating DeepSeek’s R1 model, BYD enhances its AI-powered perception capabilities to manage complex urban driving situations, from recognizing traffic signals to navigating busy intersections. The system is trained on 72 million kilometers of real-world data collected daily from the cloud and leverages a centralized architecture for fast decision-making. The DiPilot system is built on BYD's Xuanji Architecture, which includes a central processor, cloud-based AI, vehicle-side AI, and various sensor networks all connected to the DeepSeek R1 for real-time decisions.

The DiPilot incorporates AI-driven perception across three different hardware tiers, which can also receive OTA updates. This strategic design allows BYD to continuously improve the R1 model by utilizing driving data collected anonymously from all DiPilot-equipped vehicles. As more cars contribute data, this creates a beneficial cycle where more data leads to better AI performance, attracting even more customers.

Overview of Hardware Tiers

BYD is currently offering its DiPilot ADAS across three main hardware tiers:

• DiPilot 100 (Entry Tier): This level relies primarily on cameras, complemented by radar and 12 ultrasonic sensors. It supports essential Level-2 features such as highway Navigate-on-Autopilot (NOA), lane changes, and remote self-parking. BYD has planned OTA updates to add city NOA to this tier in the future.
• DiPilot 300 (Mid Tier): This includes a single LiDAR unit for enhanced 3D mapping and improved object detection capabilities. The addition of LiDAR makes city navigation more efficient and enhances obstacle detection, especially in poor visibility conditions.
• DiPilot 600 (Premium Tier): This high-end tier features multiple LiDAR sensors, providing comprehensive coverage around the vehicle. It is designed for flagship models such as the Yangwang U9 and can manage intricate urban situations and high-speed highways with enhanced reliability and 360° awareness.

Each tier is based on BYD’s Xuanji Architecture, which consolidates all sensor information into a single processing unit.

BYD, DeepSeek, ADAS