Interpretation of GAC Group's pure vision intelligent driving technology without pictures

At present, the reasons that limit the popularization of urban intelligent driving functions are mainly divided into two aspects. First, autonomous driving relies on high-precision maps. In my country, the highway network is spread all over the country, but the mileage covered by high-precision maps only accounts for about 6% of the national highways, and is mainly concentrated in highways, expressways and first- and second-tier cities. In addition, the update frequency of high-precision maps is low, and it is difficult to reflect the complex and changeable road conditions in Chinese cities in a timely manner. Therefore, it is key to develop a full-scenario autonomous driving system that no longer relies on high-precision maps. Secondly, without high-precision maps, there is no God's perspective on complex lane line information. For example, autonomous driving may not know which lane the vehicle is in, let alone which of the many signal lights matches the lane line. Therefore, the vehicle needs to have the ability to recognize and understand lane lines and road semantic information. However, lidar cannot recognize complex traffic information with rich semantics such as signal lights, traffic signs, and text.

GAC's pure vision intelligent driving road test without pictures

The GAC X Lab team developed and built a domestically exclusive pure vision-based intelligent driving system that can be used in complex and ever-changing scenarios. It has basically realized the high-speed and urban NDA functions in the core urban area of ​​Guangzhou. This is achieved without high-precision maps, and it has also verified that the camera-based autonomous driving system can drive normally in the rain and at night.

Compared with other intelligent driving systems, GAC's map-free pure vision intelligent driving system has three leading capabilities in visual recognition.

GAC's pure vision-based intelligent driving system has three leading capabilities

The first is the ability to understand complex traffic roads with extreme accuracy. Through GAC's self-developed advanced artificial intelligence algorithms based on Transformer and occupancy networks, traffic road recognition can be achieved, providing three-dimensional spatial perception for high-level intelligent driving solutions, and allowing the real world to be reproduced digitally in real time. In the test, GAC's map-free pure visual intelligent driving system can provide effective lane information in complex scenes such as on- and off-ramps, tunnels, curves, intersections, and forks, thereby helping vehicles make correct decisions.

GAC's map-free pure vision intelligent driving system can understand complex traffic roads with extreme accuracy

Second, it has a strong ability to identify traffic signals. GAC's image-free pure visual intelligent driving system uses advanced artificial intelligence algorithms to accurately identify various signal light elements, including signal light types, various symbols and colors, countdowns, and flashing states.

GAC's pure vision intelligent driving system has super traffic light recognition capabilities

The third is the ability to handle complex and ever-changing scenarios. China's traffic scene complexity is world-renowned. In order to make users feel more comfortable and safe, GAC's map-free pure visual intelligent driving system uses a large perception multi-task model based on BEV+Transformer. It has achieved a wide perception range, reaching more than 150 meters in front of the car, 100 meters behind the car, and 50 meters to the left and right, which is equivalent to the area of ​​​​three football fields, which is 1.5 times the industry level. There are more perception categories, not limited to common traffic factors such as vehicles, pedestrians, riders, and ice cream cones, and irregular obstacles can be more accurately identified. The recognition accuracy of obstacles has also reached a longitudinal error of less than 1% within 50 meters and a lateral deviation of less than 0.5%, which can easily cope with complex scenes.

GAC's pure vision intelligent driving system without images is capable of handling complex and changing scenarios

With the above powerful visual recognition algorithms, GAC's map-free pure visual intelligent driving system solves the problem of vehicle "seeing". At the same time, there are three major innovations in making decisions and executing information. The first is the real-time tracking, intention recognition and trajectory prediction of dynamic traffic objects. The second is the establishment of a command brain that coordinates data-driven and expert rules. The third is the first fluid space-time joint planning technology to solve the game between the vehicle and other traffic participants.

In addition to driving scenarios, GAC's picture-free pure visual intelligent driving system also performs well in parking scenarios - it can achieve the industry's most difficult three-dimensional parking function.

GAC's pure visual intelligent driving system without pictures

Capable of achieving the most difficult three-dimensional parking function in the industry

By utilizing large models in automatic parking scenarios, GAC's map-free pure vision intelligent driving system was the first to achieve high-precision parking based on pure vision without the use of ultrasonic radar last year. The minimum error can be controlled to 5cm, solving the pain points of users in the industry in difficult parking spaces such as three-dimensional parking spaces and narrow parking spaces.

In the field of autonomous driving technology, GAC has laid out two major technical routes: the mainstream "multi-sensor fusion technology" and the forward-looking "pure vision technology without image", both of which will get rid of the dependence on high-precision maps. GAC's goal is "where there are roads, there are GAC smart cars."

Among them, the newly released image-free pure visual intelligent driving system will be mass-produced and installed in 2026 to achieve high-level intelligent driving. In addition, this system is still continuously iterating algorithms and evolving through GAC's self-developed data closed loop and simulation system.

GAC's other urban NDA based on multi-sensor fusion has been installed on Haobo models and is gradually being opened to all Haobo users, striving to achieve "nationwide coverage" in the second half of the year.

GAC Group hopes to enable more GAC car owners to experience the convenience and fun brought by GAC Technology, and will be one step closer to its great vision of continuously creating value for a better mobile life for mankind!