What is an integrated intelligent driving solution? Huanyu Zhixing's approach is to provide you with a high-performance computing unit hardware, reliable software, and a cloud operation platform.

Written by AutoR Intelligent Driving Wang Nuoyi

In the era of artificial intelligence, all entrepreneurs want to find their place in the course of history and leave their names.

On September 17, during the 2018 Artificial Intelligence Conference held in Shanghai, Huanyu Zhixing, a new technology company focused on the research and development of connected autonomous driving systems and components, held the "Autonomous Driving Intelligent System Athena Launch Conference" in Shanghai, officially releasing the Athena software system.

Athena mainly runs on Huanyu Zhixing's self-developed Titan autonomous driving domain controller, and the combination of the two marks the official completion of Huanyu Zhixing's integrated software and hardware open platform.

Huanyu Zhixing CEO Li Ming stated, "This Athena system is aimed at improving the computing, tools, and algorithms of high-performance AI, striving to perfectly solve the applicability and reliability issues of perception and decision-making. This is our expectation for today, and we have taken a step towards perfection."

In Li Ming's vision, Huanyu Zhixing is divided into three parts: a high-performance computing unit hardware, reliable software, and a cloud operation platform.

Unlike many autonomous driving startups that prefer to use industrial control computers for computing, Huanyu Zhixing, upon its establishment, built the controller TITAN based on NVIDIA's SoC TX2, independently launching an integrated controller.

It is reported that the third generation TITAN III has been developed, which is only the size of a router, with a peak power consumption of 60W. Like the zFAS domain controller equipped in the Audi A8, TITAN III also uses Infineon's AURIX series MCU to detect the status of sensors and computing units to meet functional safety requirements.

Since the positioning is at L4 level, TITAN III can run deep learning algorithms on 6 camera channels simultaneously, running the UBUNTU operating system. From the on-site product perspective, this controller supports various hardware interfaces such as USB, CAN, RS232, Ethernet, AD/DA, etc., and is equipped with a standard TX2 board and interface board, supporting dynamic hardware expansion, allowing for an additional TX2 and interface board.

Functionally, the Titan autonomous driving domain controller supports high-precision maps, adaptive cruise control systems, lane departure warning, lane keeping systems, front collision warning, automatic lane change, intelligent human interaction, pedestrian detection systems, automatic emergency braking systems, and one-click ride-hailing.

The reliability of the lane keeping system, pedestrian detection system, and automatic emergency braking system has been well validated in the first i-VISTA autonomous vehicle challenge held in August.

The competition results show that a heavy truck from Huanyu Zhixing, measuring 17 meters long, 3.5 meters wide, and 4.5 meters high, weighing 25 tons, was able to successfully stop just 0.2 meters away from an obstacle.

On the software level, Huanyu Zhixing has achieved high-definition map collection, real-time 3D fusion perception and positioning, data storage, and high-speed Ethernet communication by integrating Athena with the TITAN autonomous driving domain controller, enabling different forms of autonomous driving functions for single vehicles, fleets, and vehicle-road collaboration, making it possible for car manufacturers to quickly validate autonomous driving technology and shorten mass production time.

To better demonstrate Huanyu Zhixing's software capabilities, the company live-streamed a demonstration of its heavy truck's 5G connected autonomous driving functions on-site, including lane keeping system, automatic lane change, one-click ride-hailing, and U-turn.

From the video live stream played on-site, it can be seen that Huanyu Zhixing's heavy truck performed well in lane keeping system, automatic lane change, one-click ride-hailing, and U-turn, but during the U-turn process, the heavy truck chose to turn in a straight lane, indicating that the vehicle still needs to learn traffic rules.

However, before testing the real vehicle on the road, Huanyu Zhixing has extensively used cutting-edge simulation technology in strategy and planning generation, which can help Huanyu Zhixing test algorithms in the conceptual stage.

According to Dr. Cao Jing from Huanyu Zhixing, "Thanks to the development of artificial intelligence technology, there have been some new changes in the field of autonomous driving, such as the transformation from past 2D image-based visual perception to current 3D visual perception."

In this regard, Huanyu Zhixing has made many breakthrough advances, including the physical description of three-dimensional information and performing deep learning directly on point clouds.

In terms of 3D perception, Huanyu Zhixing's solution includes a body feature extraction layer, extracting features from point clouds and images respectively; a feature fusion and selection module; and an optimized positioning perception module.

In the feature fusion layer, the lowest level is data-level fusion, the middle is feature-level fusion, and there is also decision-level fusion that allows sensors to make judgments about the surrounding environment independently.

For the positioning solution, Huanyu Zhixing consolidates information into infrastructure that can be constructed in advance in autonomous driving, without needing to be built in real-time, and generates road signs and road sign extraction through the fusion of image laser and point clouds.

In terms of decision planning, the planning part of the Huanyu Zhixing Athena system uses a navigation-level global planning, where the positioning module tells the current location, inputs the coordinates of the desired destination, and the high-precision map provides a complex network with topological relationships, allowing us to search for the shortest path from the starting point to the endpoint.

This is very similar to the map navigation we usually use on our phones.

In summary, Huanyu Zhixing uses a unified method based on deep learning to represent heterogeneous data features, and then employs big data methods to select features, resulting in a relatively good intelligent driving 3D value outcome with the optimized feature subset.

As an emerging technology company in the field of autonomous driving, Huanyu Zhixing began research and development related to autonomous driving as early as 2008, and has since established a series of leading autonomous driving technologies, including laser radar and camera fusion perception positioning, decision planning, simulation platforms, and intelligent connected vehicle fleets, covering both hardware and software aspects.

Academician Zhang Jianwei pointed out, "Using artificial intelligence for autonomous driving, future considerations should focus on more human-like multimodal information fusion directions beyond vision and laser radar, while also drawing on current achievements in reinforcement learning, continuous learning, and transfer learning to enhance the learning capabilities of autonomous vehicles. It is necessary to solve the 'three views' problem of autonomous driving, especially the moral view, as well as the multi-robot game for large-scale collective intelligence. The research on these key issues and the development of related technologies are also the direction of cooperation between Academician Zhang's German team and Huanyu Zhixing Research Institute."

Finally, Li Ming revealed that the integrated hardware and software solution of Huanyu Zhixing is nearing completion. The next phase will focus on how to further enhance the reliability of this open platform based on the existing foundation, as well as how to commercialize this solution.

In November, at NVIDIA's GTC China 2018 conference, Huanyu Zhixing will announce specific commercialization paths, including autonomous driving application scenarios for different vehicle types such as heavy trucks and passenger cars.