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

Article丨AutoR wisdom driving Wang Nuoyi

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

On September 17th, during the 2018 Artificial Intelligence Conference in Shanghai, the new technology company Huanyu Zhixing, which focuses on the research and development of networked autonomous driving systems and components, held the "Unmanned Intelligent System Athena Press Conference" in Shanghai and officially released Athena. Software system.

Athena mainly runs on the Titan autonomous driving domain controller independently developed by Huanyu Zhixing. The combination of the two marks the official completion of the construction of an open platform integrating Huanyu Zhixing's software and hardware.

Li Ming, CEO of Huanyu Zhixing said, "This Athena system is to improve the calculation, tools and algorithms of high-performance AI, and strive to perfectly solve the problems of applicability and reliability of perception and decision-making. This is what we expect from today. We A step towards a perfect solution."

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

Unlike many self-driving startups that prefer to use industrial PCs for computing, Huanyu Zhixing was just established and built a controller based on Nvidia's SoC TX2, TITAN, and independently launched an integrated controller.

It is reported that the third-generation TITAN Ⅲ has been developed, and its size is only equivalent to a router, and the peak power consumption is 60W. Like the zFAS domain controller on the Audi A8, TITAN Ⅲ also uses Infineon’s AURIX Series MCUs detect the status of sensors and computing units to meet functional safety requirements.

Since the positioning is at the L4 level, on TITAN Ⅲ, you can run the deep learning algorithm and the UBUNTU operating system on the 6-channel camera at the same time. From the perspective of on-site products, the controller supports USB, CAN, RS232, Ethernet, AD/DA Various hardware interface access, built-in standard TX2 board and interface board, and supports dynamic hardware expansion, you can expand an additional TX2 and interface board.

Functionally, the Titan Autopilot Domain Controller supports high-precision maps, adaptive cruise systems, lane departure warnings, lane keeping systems, front collision warnings, automatic lane changes, intelligent manual interaction, pedestrian detection systems, automatic emergency braking systems, One-click car calling.

The reliability of the lane keeping system, pedestrian detection system, and automatic emergency braking system has been well verified in the first i-VISTA Autopilot Challenge held in August.

The result of the competition showed that Huanyu Zhixing, a heavy truck with a length of 17 meters, a width of 3.5 meters and a height of 4.5 meters, and a weight of 25 tons, can successfully stop at a distance of only 0.2 meters from obstacles.

At the software level, by integrating Athena with the TITAN autonomous driving domain controller, Huanyu Zhixing realized high-definition map acquisition, real-time three-dimensional fusion perception and positioning, data storage and high-speed Ethernet communication, as well as different forms of bicycle, fleet and road collaboration. The automatic driving function makes it possible for car companies to quickly verify automatic driving technology and shorten the mass production time.

In order to better prove the software capabilities of Universal Smart, Universal Smart demonstrated its 5G networked automatic driving function of heavy trucks on the spot, including the lane keeping system, automatic lane change, one-click car calling, and U-turn.

It can be seen from the live broadcast of a live video broadcast that Huanyu Zhixing heavy trucks perform better in lane keeping systems, automatic lane changes, one-click car calls, and U-turns. However, during the U-turn process, heavy trucks choose to turn in straight lanes and vehicles are in traffic The rules still have to be learned.

However, before the actual vehicle road test, Huanyu Zhixing used a lot of cutting-edge simulation technology in the generation of strategies and plans. It can help Huanyu Zhixing test algorithms in the conceptual stage.

According to Dr. Cao Jing, Huanyu Zhixing, "Thanks to the development of artificial intelligence technology, some new changes have also occurred in the field of autonomous driving, such as the change from the visual perception around 2D images to the current 3D visual perception."

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

At the 3D perception level, Huanyu Zhixing's solution includes the physical evidence extraction layer, which extracts its features from the point cloud and the image; the feature fusion and feature selection module and the module that optimizes positioning perception in the future.

At the level of feature fusion, including the lowest level is data-level fusion, the middle is the feature-level fusion, and there is a decision-level fusion that allows the sensor to make a judgment about the surrounding environment alone.

For the positioning solution, Universal Smart Group summarizes the information into the infrastructure. This infrastructure can be built in advance in autonomous driving. It does not need to be built in real time. Through the fusion of image lasers and point clouds, road signs and road signs are generated. Generation extraction.

At the level of decision-making and planning, the planning part of Athena system of the world is through a navigation-level global planning. The positioning module tells the current position and enters the coordinates of the destination. The high-precision map will provide a complex network with topological relationships. Use this information to search for the shortest path from the beginning to the end.

This is very similar to the map navigation we usually use on mobile phones,

In summary, based on deep learning, Huanyu Zhixing uses a unified method to represent the characteristics of heterogeneous data, and then uses the method of big data to select features, and uses the optimized feature sub-levels to obtain a better intelligent driving. 3D value results.

As an emerging technology company in the field of autonomous driving, Huanyu Zhixing began research and development of autonomous driving as early as 2008. So far, it has established a combination of lidar and camera fusion perception positioning, decision planning, simulation platform, intelligent networked fleet, etc. Including a series of leading autonomous driving technology, covering two different aspects of hardware and software.

Academician Zhang Jianwei pointed out, "With artificial intelligence for autonomous driving, in the future, we should focus on considering the direction of multi-modal information fusion of more humans than vision and lidar. At the same time, we can learn from the current results of reinforcement learning, continuous learning, and transfer learning. In order to enhance the learning ability of autonomous vehicles, it is necessary to solve the "three views" of autonomous driving, especially morality, and multi-robot games for large-scale group intelligence. Research on these key issues and research and development of related technologies are also academician Zhang The direction of the German team and Huanyu Zhixing Research Institute in cooperation."

Lastly, Li Ming revealed that the current software and hardware integration plan of Huanyu Zhixing is nearing completion. The focus of the next stage will be on how to further improve the reliability of the open platform on the basis of the existing basis and let this solution be carried out. Commercialization.

In November, at Nvidia’s GTC China 2018 conference, Universal Smart will release specific commercialization paths, including automatic driving application scenarios for different models such as heavy trucks and passenger cars.