L3 Intelligent Driving Says Goodbye to "Night Blindness": The "Deeply Understanding the Future" AI Imaging Engine Has Been Applied in GAC | Early-stage Project

Author: Ou Xue

Editor: Yuan Silai

Yingke has learned that the Complex Light AI Computational Imaging Engine of "Deeply Understanding the Future" has been mass - produced and launched with GAC's high - end flagship model, the Haobo HL. In the future, it is expected to be extended to more than a dozen different GAC models.

Deeply Understanding the Future was founded in 2017. Its main technological barrier lies in the AI ISP technology system based on the full - link neural network, which enables ordinary visible - light vision sensors to present full - color, high - definition, and real - time images at night.

Overall, the company has currently established three major product lines: the Guangyu series - UAV payloads, the Polaris series - handheld imaging devices, and the Tieshi series - module products. The company's first product was officially launched in July 2022, and it has achieved a high - speed sales growth of over 200% for three consecutive years.

The shooting effect of the handheld camera of Deeply Understanding the Future in a dark environment (Photographed by Ou Xue)

Previously, Deeply Understanding the Future had successfully launched products such as mounted night - vision cameras for industrial drones, handheld night - vision cameras, and observation glasses. Currently, the automotive field has become the company's most important strategic growth engine, and the company has cooperated with several domestic first - tier automakers.

"2025 is the Cambrian period of intelligent driving, with various vehicle intelligent technologies booming. Smart cars will become the largest carrier of future AI algorithms. The threshold to enter this field is extremely high, but once entered, the company's growth potential will far exceed that of other fields." Zhang Qining, the founder of Deeply Understanding the Future, told Yingke. After the relevant release with GAC Group, many mainstream automakers have actively contacted the company to seek cooperation.

Zhang Qining further stated that perception, decision - making, and control are the three elements of an intelligent agent. In the development of intelligent driving, with the maturity of decision - making and control, the accuracy and stability of perception have become the bottleneck of the intelligent system. The AI imaging engine capable of adapting to complex light will see an explosion in the field of perception and will be closely coupled with intelligent driving algorithms on the domain controller and undergo rapid OTA iterations synchronously.

"Autopilot has reached the L3 level, but in - vehicle cameras are still at the L1 level." Zhang Qining explained that as the most important "eyes" of autopilot, the price of cameras has remained in the range of less than 200 yuan. The imaging adaptability is far lower than that of human eyes, which exposes intelligent driving to safety risks at night and in complex weather conditions due to the inability of cameras to form clear images, remaining at the L1 level.

According to Zhang Qining, traditional ISP algorithms are based on artificial rules, resulting in poor image quality in complex light scenarios. The new - generation AI computational imaging simulates the calculation method of the human visual neural network and can iteratively learn and continuously evolve according to the requirements of intelligent driving algorithms.

Comparison chart of two algorithms (Image source: Deeply Understanding the Future)

AI ISP is a field with a low entry threshold but an extremely high threshold to "do well". Deeply Understanding the Future is currently the only artificial intelligence enterprise globally that has achieved a commercial closed - loop of a vehicle - grade full - process neural network - based AI imaging engine.

From a technical perspective, Zhang Qining said that the current AI imaging neural network architecture is still in the process of iteration, and there is still a relatively large room for algorithm development. It has not reached a stage suitable for chip implementation, and solidification at this time is a hindrance to algorithm development.

"Thanks to the significant growth of business in the automotive field, the company is expected to become the first artificial intelligence imaging company to achieve profitability in the AI ISP field this year." Zhang Qining said.