The battle of autonomous driving: single-vehicle intelligence vs. intelligent networking

The emergence of advanced driver assistance systems has brought autonomous driving closer and closer to us. As autonomous driving technology matures and commercialization accelerates, cars are no longer just a means of transportation, but are gradually transforming into diversified smart entertainment hardware. When consumers buy cars, safety is not only the only indicator, but comfort and entertainment are also factors to consider when buying cars. In the future, the emergence of fully autonomous driving will promote another transformation in the automotive consumer market, and autonomous vehicles will change people's travel habits.

At this stage, autonomous driving technology is not yet mature, and there are still many possibilities for the development of autonomous driving. Among them, single-vehicle intelligence and intelligent networking are the two major technical routes for the development of autonomous driving technology.

The intelligence of a single vehicle mainly relies on sensors such as millimeter-wave radar, lidar, on-board visual cameras, wire control systems, and computing unit hardware carried by the vehicle for environmental perception, decision-making, control, and execution, allowing the vehicle to achieve independent thinking and decision-making like humans, thereby driving the vehicle to the predetermined destination.

Intelligent networking is to connect the elements of "people-vehicle-road-cloud" through the Internet of Vehicles technology, so as to upgrade the vehicle's environmental perception, decision-making, control and execution functions, enhance the autonomous driving function, promote the management and improvement of the transportation network, thereby providing a safer, more comfortable, more energy-efficient and more environmentally friendly driving method, promote the regulation of urban transportation systems, and build a new type of smart city.

Bicycle Intelligence

The development model of single-vehicle intelligence mainly relies on the coordination of hardware equipment and software installed on the vehicle to realize the vehicle's automatic driving function. This development model of single-vehicle intelligence has higher requirements for environmental perception. In order to make single-vehicle intelligence safer, it is necessary to install enough hardware equipment on the vehicle to eliminate the visual blind spots during the driving of single-vehicle intelligent vehicles. However, after so long of development, single-vehicle intelligence has not been able to achieve large-scale commercial use. The main reason is that it is difficult for single-vehicle intelligence to achieve sufficiently high safety.

In order to obtain better perception effects, it is necessary to install on-board visual cameras, ultrasonic radars, millimeter-wave radars and other hardware equipment on the vehicle. However, in actual use, there are still blind spots, and it is impossible to achieve all-round recognition. In addition, the road conditions are complicated, and obstacles behind the roadside cannot be recognized by these hardware. If there is an object behind the obstacle that prompts the obstacle to move, and the vehicle does not make a prediction, it will lead to more dangers. In addition, the installation location, field of view, data throughput, calibration accuracy, and time synchronization requirements of the perception hardware equipment are also very high. When a single-vehicle intelligent vehicle is driving in busy intersections, bad weather, small object perception and recognition, signal light recognition, backlight and other environmental conditions, it is still difficult to completely solve the problem of accurate perception and recognition and high-precision positioning.

In addition, the sensing range of the on-board visual cameras, ultrasonic radars, and millimeter-wave radars currently installed on vehicles is very limited, and the safe driving distance of vehicles needs to be detailed to the centimeter level or even the millimeter level. If there are vehicles on the road that are driving very fast and the vehicle does not make a timely prediction, it will cause the vehicle to be unable to respond in time, leading to danger. Many major car manufacturers now provide L2 and L3 level autonomous driving. When encountering such a situation, the driver needs to take control of the vehicle in time. If encountering similar problems, the driver will not be able to take over and respond in time.

The development of single-vehicle intelligence is inseparable from high-precision maps. The collection of high-precision maps is very complex and costly. If the road is rectified or repaired, the high-precision map needs to be updated in a timely manner. If the route of single-vehicle intelligent autonomous driving is commercialized, the high-precision map needs to be updated at any time according to road conditions. The update frequency requirement is even higher, which cannot be achieved at the current stage.

Excessive cost is also one of the main reasons that hinder the development of single-vehicle intelligence. Due to the high cost of hardware equipment, in order to process the signals generated by these hardware devices, computing units need to be configured, and corresponding software systems will be deployed on the vehicle side. The maintenance, management and update of the software system also greatly increase the maintenance cost of autonomous vehicles. At present, the hardware equipment of L4 autonomous driving vehicles generally includes: 6 to 12 on-board visual cameras, 3 to 12 millimeter-wave radars, 5 or less laser radars, 1 to 2 GNSS/IMUs and 1 to 2 computing platforms. The total cost is very high, which makes the commercialization of single-vehicle intelligence more difficult.

There are currently two feasible solutions for bicycle intelligence.

One is the solution represented by Google, which equips vehicles with perception hardware such as lidar, millimeter-wave radar, and on-board visual cameras. The disadvantage of this solution is that it is too expensive and the possibility of commercialization is low, but the coordination of multiple perception hardware also increases the safety of autonomous driving, which is sought after by many major car manufacturers. It is also the main solution recognized by many OEMs at this stage.

The other is a solution represented by Tesla, which achieves autonomous driving through the collaboration of on-board visual cameras and software. This solution is relatively low-cost and has a high possibility of commercialization. At this stage, we have seen the prototype of commercialization, but autonomous driving is only based on on-board visual cameras, with fewer information sources and insufficient security guarantees.

Although the technologies used in the two solutions are very different, overall, they both take the route of single-bike intelligence.

Intelligent Networking

As the development of single-vehicle intelligent technology reaches a bottleneck, intelligent networking has gradually been mentioned. Intelligent networking refers to the organic combination of vehicle networking and single-vehicle intelligence, integrating modern communication and network technologies into single-vehicle intelligent technology to achieve information interaction and sharing between vehicles, vehicles and people, vehicles and roads, vehicles and background (V2X), etc., and realize information communication between vehicle driving information and the main vehicle manufacturer, management department, and destination facilities, promote the realization of safe, comfortable, energy-saving, and efficient driving solutions, and ultimately realize the commercialization of fully automatic driving.

The technical solution of intelligent networking is to make vehicles smarter and then make vehicles and road facilities smarter, from single-vehicle intelligence to group intelligence. Since the commercialization cost of single-vehicle intelligence is too high, using roadside equipment to replace some technologies and making road facilities smarter can effectively reduce the manufacturing and R&D costs of vehicles and realize the possibility of commercialization as soon as possible. The premise of intelligent networking is the intelligent transformation of road facilities and investment in infrastructure, which requires participation from many parties. For example, whether government management departments will support the intelligent transformation of road facilities, whether major car manufacturers are willing to share vehicle driving information, whether investors will support such technical solutions, etc., will affect the development and evolution of intelligent networking technology.

The development of intelligent networking cannot be achieved overnight. It needs to be realized step by step. It is mainly divided into three stages. The first stage is the information interaction and coordination stage, which can realize the information interaction and sharing between vehicles and roads, and realize functions such as collision warning and road hazard prompts. The second stage is the collaborative perception stage. On the basis of the first stage, the perception and positioning of vehicles and road facilities are realized. The third stage is the collaborative decision-making stage. On the basis of collaborative perception and positioning, the driving decision and control of the vehicle are realized, so as to realize the possibility of fully automatic driving. The realization of intelligent networking allows vehicles to collect road information in advance during driving, so as to predict driving behavior.

my country's research and development in intelligent networking is rapid. In the first stage, large-scale testing and verification and pilot demonstration have been carried out in many cities, and commercial operations have been gradually carried out on a trial basis. In the second stage, construction, deployment and testing have been carried out in some cities. The collaborative perception capability of roads has been fully verified, and the coordination and control of infrastructure in some scenarios have also been explored. However, the current technological development is still unable to achieve the commercialization of fully autonomous driving.