You are here : Home News Press Center Text
After having driven the test vehicle, we visited the principal of NavInfo's automated driving team to learn more about the technical details.

 On November 16, in a parking lot not far from the venue of NavInfo"s annual User"s Conference, a test vehicle of automated driving independently researched and developed by NavInfo was set out. At the same time, Cheng Peng, CEO of NavInfo, announced in a site several kilometers away that NavInfo was about to establish the new strategic layout of the "intelligent automobile brain".

It was only about one year"s time from the point in the last year when NavInfo gradually revealed its transformation plan orientated to automated driving in public occasions.

If we review the pan-automated driving market today, we can find that it has become pretty intensively with many players in the industry, which include but are not limited to automobile manufacturers, Tier 1, technology innovation enterprises, Internet companies, etc. However, in terms of an enterprise which starts business in navigation maps and then turns the business course to automated driving solutions, what would be different?

In order to figure out this question, after having participated in the official test driving activity organized by NavInfo for their automated driving vehicles, · AI-Drive paid another visit to Ma Zhou, the principal of NavInfo"s autonomous driving team, to seek the answer.

Speaking from the testing vehicle

On the very day that we participated in the open test driving organized by NavInfo, the weather in Beijing was quite cold with winds. When we arrived in the test yard, there was a parking lot not far from the venue of the User"s Conference convened by NavInfo, a blue test vehicle modified from Haval H7 for automated driving was already there.

Under the current background of quite diversified sensor configuration schemes, this test vehicle of NavInfo appears to be much simpler. The equipment for environment perception only covers an 8-line laser radar and a single camera.

After the initial view, · AI-Drive joined the subsequent test ride in the vehicle. Sitting in the back seat, we didn"t find complex circuits and redundant computerized devices. After we took seats, the driver activated the intra-vehicle interaction system developed by NavInfo, by simply saying "Hello, Xiaoxin", and vehicle starting and braking were both realized subsequently by audio interaction.

Within nearly one minute after startup, the vehicle showed movements like steering, obstacle avoidance planning, parking, etc., with an average speed of 30-40km/h and a straight driving distance up to 70m. According to Ma Zhou, the vehicle was covered by the 4G network, and autonomous vehicle calling with no persons in the vehicle can be realized, other than the demonstrated automated driving with the driver and passengers inside the vehicle.

As explained by Mazhou, thanks to the high-accuracy map data (HAD) with traffic rule information (e.g. left-turn roadway, straight roadway, etc.) defined by NavInfo, the automated driving system has converged a lot of road and environmental attributes - "Map is the abbreviation of traffic rules" - in this way, the environment sensors can focus on relatively dynamic road environments, e.g. detection of persons, vehicles and obstacles, etc., which significantly reduces the overwhelming dependency of the system on environment sensors. Since it was constrained by the current regulations, these more complex planning calculations were not completely presented.

"Map is a huge chessboard"

It was in 2016 that people started to hear about the news of NavInfo"s transformation to automated driving. However, Ma Zhou told · AI-Drive that NavInfo"s initial start in automated driving was even earlier.

As early as 2015, NavInfo had started internally to build one R & D team for automated driving, and certain foundation work had been accomplished by the end of 2015. In April 2016, the team started to build the framework of the automated driving system and arrange more preparation work. In August 2016, R & D of the finished automobile was initialized. Thereafter, NavInfo officially announced the Automobile Automated Driving Project Cooperation Agreement entered into with the Great Wall Motor in December 2016, and declared to "jointly develop automated driving functions based on automobile models in mass production". At that time, a test vehicle developed by NavInfo"s automated driving team based on Great Wall Motor had been running on the road.

Currently, as the principal of NavInfo"s automated driving team, Ma Zhou will lead the team to upgrade their finished automobiles in terms of overall sensing, decision-making and control of the automated driving system, and much of the work has been systematized, in contrast with their exploration of initial establishment started from scratch.

Though this systematization is reflected in the integration of the finished automobiles, what more distinguished is actually the driving decision-making brain featured with "predictability" and constructed from the foundation of NavInfo"s proprietary navigation maps and high precision map data.

In order to explain this feature intuitively, Ma Zhou took the Chinese chess as an example.

Take the rule of " the chessman ‘horse"s going diagonally " in Chinese chess as example, when the computer is directly told to let " the horse go diagonally”, it is impossible for the computer to make decisions to do that. However, if this rule is put in the environment of a chessboard, it is much easier to realize. As the action of "going diagonally" can be broken down to: the chessman "Horse" can move two grids through going left, right, up or down in one direction, and then move one grid through going either left or right. In this way, a chessman can explore all possible routes from any direction. Of course, in Chinese chess, there is also a rule of "Obstacle". If there is an obstacle in the first step, stop exploring it.

The above steps are a process used to understand the rule of" the chessman ‘horse"s going diagonally " from the perspective of the computer algorithm, and it can provide the most critical information for the research and development of automated driving, which is summarized by Ma Zhou into a simple sentence -

"Map is to provide a huge chess board for the route decision-making of automated driving automobiles."

Of course, if the rule of" the chessman ‘horse"s going diagonally " is the constraint on the movements of the chessman "Horse" in the Chinese chess, then the traffic itself is also a convergence of vehicle driving rules. For instance, the number of roadways of a road, the forward direction of a roadway, the indicated direction of a traffic light, and many other similar attributes. What NavInfo focuses on is to empirically incorporate this information into high precision maps, allow the rules to instruct automated driving of the vehicles in advance, and reduce reprocessing procedures after the identification of these rules by sensors such as cameras.

The objective is to provide an automated driving integration scheme centering in decision-making

Ma Zhou told (Official account: · AI-Drive that, technically, what NavInfo plans to deliver is an automated driving scheme centering in decision-making. This is because, in whatever a traffic environment, a vehicle must be aware of "Where it is, where the traffic lamps are, what vehicles are around and where they are going to", and these reference information can be provided by the decision-making instructions based on high precision map data enabled by NavInfo.

"Everyone is focusing on the point they are best at," Ma Zhou said.

Amongst all the players entering the industry of automated driving nowadays, some may be from the automobile industry, some may be sophisticated in automobile control, and some may be from the technology circles such as the Internet circle and good at environment perception schemes based on artificial intelligence. From this perspective, the technical path selected by NavInfo is actually logical.

About two or three years ago, NavInfo put forward the high precision map defined by its own standard and applied to HAD (Highly Automated Driving) for the first time. After that, NavInfo and the auto plant made clear use scenarios for the development and use of different specifications. At the User"s Conference of NavInfo this year, Cheng Peng announced that HAD high precision map entered the stage of "preparation for mass production", and at that time, HAD covered 70,000 km of national highways with the accuracy of 20cm, supporting periodic updates.

Cheng Peng, CEO of NavInfo, introduced the HAD map at the 2017 User"s Conference

On the other hand, NavInfo has currently been connected to multi-source position data, including DiDi, which can update real-time dynamic traffic information, such as temporary road closures and accidents. All of these data will serve driving decision-making via cloud for automated driving cars.

So much for automated driving, here is an advertisement. On January 16, 2018, · AI-Drive will hold GAIR Silicon Valley - The Future of Autonomous Driving in China and U.S. in the Silicon Valley, the science and technology base camp in the United States. At this summit, dozens of top automated driving teams from China and U.S. will deliver speeches, and in addition to that, nearly 100 automated driving industry chain companies will exchange at the scene. At present, booking of Christmas discount tickets is still open to the public. For more details, please visit

Regularization + non-regularization with integration of two-layer driving brain

Ma Zhou told · AI-Drive, strictly speaking, the current automated driving achieved by NavInfo is the L3 level, and a description of the concept comes from the fact that the automated driving ability of NavInfo can be fully achieved on highways except toll stations, service areas, including autonomous upper and lower ramps, etc.

The above-mentioned L3 driving ability is realized based on the "dual integration decision-making" built by the automated driving team of NavInfo. To be specific, first integrate sensor and map with the introduction of traffic rules and other data; then, integrate identification information of the sensor with maps to avoid obstacles.

From the perspective of NavInfo, no matter what kind of sensor configuration and automated driving schemes are, integration with map data is very important. For example, the attribute of the specific position of each traffic light is marked in the high precision map data. If the camera sensor mistakes the high-mount stop lamp on the front car for the red light, at this time, the integrated high-precision map data can remove this error.

Ma Zhou introduced that two integrated operations of NavInfo came from the two-layer "driving brain" of the automated driving vehicle respectively with one layer being regularized, and the other non-regularized.

Regularization refers to a map-based driving brain, which can solve almost 90% of the driving problems;

However, in fact, not every driver obeys the rules, so in some cases, the irregular driving brain is required to play a role.

Ma Zhou took a simple scene as an example. For example, rules require vehicles to move in roadway lines, but if only half a roadway can be used due to driving on the wrong side of the road or an accident, what would you do? At this time, you need to think out of the box and allow half-roadway driving. In this part, deep learning needs to be introduced to allow vehicles to understand when a similar decision needs to be made.

Initial product

Now, according to Ma Zhou, NavInfo has signed an order with customers demanding automated driving schemes, and more work is in progress. However, before the more ideal high-end automated driving product, how can NavInfo find the initial implementation for results at the present stage?

As for this question, Ma Zhou told · AI-Drive that automated driving cannot form products of mass production in public environments at this stage, but NavInfo is looking for the implementation of the automated driving technology in specific scenarios from the perspective of two directions: one is the specific area, and the other refers to the specific function.

So much for specific areas, logistics and other scenarios belong to this category; as for specific functions, it"s actually a good choice for the automated driving industry to realize products on public roads in the early period. Ma Zhou used "predictive and active safety products" to describe attempts of products made by NavInfo in this field.

For a simple example, if cruise control of a car is set up to 100km/h, it will travel at this speed even in terms of cornering, and this often requires the driver to intervene to slow down. The scheme of NavInfo is based on road parameters in the high precision map, such as the slope, camber, etc. to determine the most appropriate speed in advance and intervene in the car"s throttle and brake to help adjustment. In addition to safety, this function can also help vehicles to reduce fuel consumption, and according to previous official release by NavInfo, this high precision ADAS-oriented map data and supporting algorithms can help save an average of 5 percent of fuel for the car engine.

Ma Zhou said that in addition to the decision-based automated driving scheme, NavInfo should be familiar with various links in the automated driving scheme through R & D of the entire vehicle to meet different needs of users.

As he said, the positioning of NavInfo has always been very clear. Whether it is the digital map provider in the past or service provider today, "the goal of our services is still auto enterprises".


Back Top