The whole process optimization from perception to execution, the picture shows the T928 assisted unmanned driving system

Self-driving, also known as driverless driving, self-driving cars do not require human action to perceive the environment. To complete independent driving, it is necessary to read the road traffic situation, analyze the road surface situation, and adjust the driving route and driving operation in time according to the road analysis results.

Autonomous driving is an autonomous vehicle that integrates perception, decision-making and execution, and its three points are also important technical elements for the landing of autonomous driving.

Road condition awareness

To complete automatic driving, seeing the road is the first requirement, similar to people walking through the "eyes" and "ears" to know the traffic situation, self-driving cars through the picture of the T928 edge computer equipped with a four-in-one interface (support up to 16 GMSL1/GMSL2 cameras), and various rail-level cameras, millimeter wave radar, and lidar sensors and other hardware devices to connect.

Through the connected hardware devices to perceive distance, direction, road condition images and other information, at the same time the obtained information through the T928 edge computer processing, and finally use intelligent network technology to communicate between cars and people, cars and vehicles, vehicles and service platforms, to help self-driving cars obtain more traffic information and ensure safer driving.

Driving decisions

If the perception system is like "eyes" and "ears", then the decision-making system is the "brain", and the information perceived by the hardware device needs to be uploaded to the decision-making system for processing. The amount of data that driverless cars need to process is huge, generating more than 1GB of data per second, and even sending some of the data to the cloud for analysis is impractical.

Based on NVIDIA Jetson AGX Orin, the picture is the T928 edge computer, the highest computing power can reach 275T, equipped with 12-core Arm Cortex A78AE v8.2 64-bit CPU 3MB L2+6MB L3, and equipped with 64 TensorCore 2048-core NVIDI Ampere architecture GPU, through T928 edge computing, the data of a large number of sensors in self-driving cars can be used. Real-time processing is performed on the data source side, and the real-time processing, powerful computing power, model reasoning and logic rules of the T928 edge computer are used to accelerate decision-making, reduce latency, and ensure driving safety.

Action execution

Executing this link is the most intuitive step of autonomous driving, playing the role of "hands" and "feet". At the end execution level of automatic driving, the vehicle needs to independently execute the instructions given by the T928 edge computer for road condition information, which is specifically reflected in the control of gear shifting, braking, steering, avoidance and other operations.

Through the figure of the T928 edge computer, the whole process of the unmanned driving system from perception, decision-making to execution is optimized, and the automatic driving function of the vehicle in high-speed and urban scenarios is realized, providing users with complete, customized and open solutions while ensuring the quality of experience and driving safety of autonomous vehicle users.