Text/BY
“Lidar is too stupid, and whoever relies on it will be doomed.” Musk has dissed lidar on many occasions. In his opinion, As long as computer vision technology is mature enough, autonomous driving can be achieved with just cameras. In fact, not only Musk, but also many autonomous driving companies have the same idea.
At this CES conference, Intel showed an autonomous driving video developed by Mobileye, which caused a burst of exclamations at the scene. The particularity of this self-driving demonstration is that the demonstration vehicle is only equipped with 12 cameras and does not use any radar! Equipped with two of Mobileye's latest Eye?Q5 chips, the demonstration vehicle achieved autonomous driving capabilities that were not inferior to Waymo's demonstration vehicles, including driving through complex intersections without traffic lights, automatically planning roads, automatically avoiding pedestrians and other difficult actions.
Mobileye said that its technical principle is to perform deep learning processing on 2D images captured by multiple cameras to generate a 3D model, thereby providing the required environmental information for autonomous driving decisions. In other words, Mobileye achieves the same function of lidar through low-priced camera and software technology.
In fact, if the algorithm and computing power are powerful enough, it is possible to "upgrade" the 2D images output by multiple cameras into 3D images. After all, that is what people do. Through the powerful computing power of the human brain, humans can generate 3D images from the 2D images "photographed" by two eyes. If humans can do this, it is only a matter of time before machines can do it as well.
In addition to using cameras to realize 3D images of lidar, Mobileye also used cameras to build high-precision maps at this CES. Mobileye has signed a cooperation agreement with SAIC to apply Mobileye's road network asset management technology REM to cars produced by SAIC through SAIC's subsidiary Zhonghaiting (a high-precision map service provider).
This technology allows the vehicle to collect road data through its own camera while driving. The data will be encrypted and uploaded to the cloud, which can eventually form a crowdsourced "road book" with updated facts to feed back the system. cars to improve their assisted driving capabilities. This is similar to high-precision maps. At this stage, high-precision maps are mainly generated through fleets equipped with lidar for data collection, and subcontracting is also a future solution that many high-precision map manufacturers are looking forward to.
Mobileye said that the cooperation with SAIC will create a road network management system that can be used by L2 and higher level autonomous vehicles. In addition, the surveying and mapping solution is an open platform that will provide support for other car manufacturers’ layout in domestic road network management-related fields. SAIC Motor will work with Mobileye's other global car company partners to collect road data to make global real-time high-precision maps possible.
For L2 assisted driving and L4 and above autonomous driving, cost will be one of the biggest obstacles to popularization. Algorithms similar to Mobileye's can make up for sensor deficiencies, and even obtain autonomous driving synergy benefit solutions through open cooperation, which will effectively reduce the cost of autonomous driving and accelerate the popularization of high-level assisted driving and autonomous driving. I think accelerating the popularization of autonomous driving is one of the themes of this year's CES. In addition to Mobileye, other manufacturers are also exploring in different directions. Please stay tuned for other reports brought by Autolab at CES? 2020!
This article comes from the author of Autohome Chejiahao and does not represent the views and positions of Autohome.