Internet of Vehicles Forum | When will vehicle-road collaborative autonomous driving arrive?

01. What can vehicle-road collaboration do for autonomous driving?

(1) The first thing that vehicle-road collaboration can do to help autonomous driving is more ubiquitous perception.

Autonomous driving has been paying a huge price to obtain accurate information about the surrounding environment. According to a survey by French technology and business market research company Yole, the current perception cost accounts for 40% of the total cost of an autonomous driving vehicle; even if the perception cost will drop significantly with technological advances, it will remain at 26% by 2032.

In addition to the high cost, the ability to perceive has always been a bottleneck restricting the implementation of autonomous driving. Whether it is a camera, millimeter-wave radar or lidar, the sensors currently used in autonomous driving are all visual sensors, and the direction of improvement has always been to see farther (increased perception distance), see wider (expanded field of view), and see more clearly (improved resolution). These indicators restrict each other and are difficult to meet at the same time. Therefore, autonomous driving has to use multiple different types of sensors and combine them with each other in order to obtain more complete coverage of the vehicle body. Multi-sensor fusion has brought about a series of calibration and spatiotemporal benchmark synchronization problems, further increasing the complexity of the autonomous driving algorithm.

Even if these problems are perfectly solved, the current maximum perception range of these sensors is only 200 meters (at the expense of resolution and field of view), which is sufficient for low- and medium-speed autonomous driving, but there are still major risks for high-speed autonomous driving.

Another challenge of quasi-visual perception is occlusion. Rain, snow, fog, people, cars, buildings, and vegetation, as long as there is occlusion, it will bring huge obstacles to quasi-visual perception. Therefore, "ghost poking" has always been the biggest challenge for autonomous driving (including ADAS systems). The aforementioned problems are mainly related to the perception of the surrounding environment and traffic participants. In addition, the perception of road traffic signs and markings in multiple scenes and multiple working conditions is not an easy task for single-vehicle intelligence. There are many types of signal lights. Strong light and backlight cause the camera recognition accuracy to decrease. Signs are damaged, and new and old road markings coexist. These all bring huge challenges to autonomous driving vehicles in obtaining information. These challenges are all areas where vehicle-road collaborative technology can play a good role.

First of all, roadside perception "stands high and sees everything". By reasonably arranging the positions of sensors, blind spots can be eliminated to the greatest extent.

Secondly, the roadside sensing position is fixed, and there is enough prior information to assist perception, thereby improving perception accuracy. Third, roadside perception can be combined into a ubiquitous perception network through advanced communication technology, which can not only notify the vehicle of near-field environmental information, but also notify the far-field environmental information hundreds of meters away, greatly improving the vehicle's perception range. Fourth, a lot of roadside information, such as traffic light status, sign content, and marking position, can be digitized, so it has very high anti-interference and can be transmitted to the car without loss. Finally, the cost of vehicle-road collaboration lies mainly in infrastructure transformation. As the number of autonomous driving vehicles receiving services increases, the cost allocated to each vehicle will be reduced to a very low level.

(2) Another important significance of vehicle-road collaboration for autonomous driving is to help autonomous vehicles interact with each other’s driving intentions. The route planning of each vehicle is a game between itself and surrounding traffic participants.

When traffic is sparse and driving routes are relatively fixed, this game is relatively simple. As long as autonomous vehicles adopt relatively conservative driving strategies, accidents can be effectively avoided. As vehicle density increases and the environment becomes more complex, autonomous vehicles are prone to being stuck in a dilemma. The automatic congestion following function is a technology that many mass-produced vehicles have adopted, but in actual use, the user experience is not ideal. Because the automatic congestion following function often requires a large distance between vehicles, the possibility of being squeezed in is greatly increased. In order to avoid rear-end collisions caused by squeezing in, drivers must always be alert and tired.

In addition to longitudinal following, lateral lane changes are more challenging for autonomous driving. Entering and exiting roundabouts, entering merging areas, overtaking, and changing lanes all require accurate prediction of the driving trajectory of surrounding vehicles and judging the possibility of conflict between the two vehicles. For human drivers, turn signals, gestures, and observations of previous driving behaviors can all help to judge the vehicle's driving trajectory. For autonomous vehicles, it is very difficult to judge the driving intentions of surrounding vehicles. The use of vehicle-road collaboration (mainly V2V vehicle-to-vehicle collaboration here) can help autonomous vehicles obtain this information. Vehicle-road collaboration can also provide a driving route negotiation mechanism for autonomous vehicles.

When the traffic volume at an intersection is small, a simple avoidance strategy can be used to avoid collisions, but avoidance will inevitably affect the efficiency of the intersection. Effective control of driving speed and driving intervals allows vehicles in different directions to pass through the intersection at a uniform speed without stopping, which can greatly optimize the regional traffic volume, and this can only be achieved by relying on autonomous driving and vehicle-road collaboration. The collaboration between autonomous driving vehicles can be divided into two-vehicle collaboration, multi-vehicle collaboration, and fleet collaboration.

Two-vehicle collaboration refers to the negotiation of driving routes through the interaction of driving intentions between two vehicles in scenes such as intersections, roundabouts, merging areas, and overtaking, so as to reduce stopping and starting, increase unit traffic volume, reduce path conflicts, and reduce collision risks. Multi-vehicle collaboration refers to the control of vehicle spacing and speed under different traffic flows to maximize road utilization. Multi-vehicle collaboration can also achieve emergency vehicle giving way in congested environments and allocation of tidal road rights, and reorganize traffic flows under the premise that road resources remain unchanged. Fleet collaboration refers to controlling the vehicle spacing and separation vectors within the fleet to maintain a compact group, reduce fuel consumption, reduce the control of the following vehicles, and save labor costs.

02. What is the significance of autonomous driving for vehicle-road collaboration?

Vehicle-road collaboration can accelerate the implementation of autonomous driving, and the implementation of autonomous driving can also help maximize the value of vehicle-road collaboration.

Autonomous vehicles can provide richer and more accurate information for vehicle-road collaboration. Autonomous vehicles are equipped with a variety of vehicle-side sensors and have high-precision positioning capabilities. Therefore, they can serve as the floating vehicle perception front end of the intelligent transportation system and provide perception information supplement for the vehicle-road collaboration system.

Autonomous vehicles can also report their abnormal status in real time, which is convenient for background intervention and avoidance of surrounding vehicles. Autonomous vehicles can provide easy-to-predict vehicle behavior. Whether it is local traffic coordination or global traffic induction, the vehicle-road cooperative system needs to know the current status of individual vehicles, route planning, and driving destination.

The more timely and accurate the information is, the better the implementation effect of the traffic management function of vehicle-road collaboration will be. The route selection of human drivers is still somewhat random, while the route of autonomous vehicles is completely obtainable. Autonomous vehicles can better abide by the rules. The previous article describes the dilemma faced by the congestion automatic following function in the mixed driving stage of human driving and autonomous driving. This is because autonomous vehicles tend to abide by established rules, while human drivers tend to look for "opportunities" to break the established rules.

When the penetration rate of autonomous vehicles is high enough, both conventional rules and rules temporarily issued by the vehicle-road cooperative system can be well implemented, thereby achieving global efficiency optimization, and each driving individual can also obtain a relatively fair right of way. Autonomous vehicles do not require complex HMI (human-machine interface) design. The vehicle-road cooperative system will eventually act on the vehicle. For manned driving, information must be optimally designed through a reasonable HMI. The method, location, content, frequency of information notification, and priority processing when information conflicts will affect the driver's information acquisition effect, so special design is required. For autonomous vehicles, this problem can be effectively avoided by simply providing the original information to the on-board controller for centralized decision-making.