Pony.ai has announced that its Level 4 (L4) automotive-grade autonomous driving (AD) domain controller has surpassed two million kilometers of on-road testing. Used in the company’s seventh-generation robotaxi fleet, the system has demonstrated high hardware reliability and software adaptability, meeting the requirements of Level 4 autonomous driving, the company says.
Pony.ai’s seventh-generation AD system is designed for a lifecycle of 10 years or 600,000 km. Through algorithm optimization and system engineering, Pony.ai has reduced the domain controller’s cost by 80% compared to the previous generation while increasing its lifespan, effectively lowering the system’s total lifecycle cost.
The L4 domain controller features three Nvidia OrinX chips in the main unit and a fourth dedicated to system redundancy, providing a total of 1,016 TOPS of computing power. Developed through a co-designed hardware-software architecture, Pony.ai custom-built the controller to meet the full system and functional requirements of fully driverless robotaxis, as well as the performance and safety demands of L4 autonomous driving algorithms. As a result, the controller has achieved a 50% to 80% reduction in size, weight, power consumption and cost.
Functioning as the central computing hub, the controller integrates system power management, interactive display, hazard warning lights, gateways and global navigation satellite system (GNSS) capabilities essential for robotaxi operation. By integrating these features into a single unit, Pony.ai has reduced wiring complexity and the number of individual components. The domain controller supports both liquid cooling and passive cooling.
The system incorporates a multi-layered safety architecture and degradation strategies, delivering fail operational capability – in the event of a primary system failure, the controller can seamlessly switch to the redundant system or the minimum risk condition controller (MRCC) to ensure safe vehicle control. Even if the main system’s power or chassis communication fails, the redundant system can still maintain critical perception (including blind spot coverage) and safe driving capabilities. This enables the vehicle to navigate intersections or ramps and safely pull over, minimizing the risk of traffic disruption or collisions.
In related news, Waabi recently launched an alternative to closed-course track testing – mixed reality testing (MRT). Read the full story here