Autonomous driving technology is an important factor affecting the future development of the automobile industry. With the maturity of autonomous driving and AI technology, cars will no longer be a driving tool subordinate to people. The core value components of cars will shift from transmission systems that embody power and operating systems to intelligent software systems. With autonomous driving, the hands, feet, and eyes of the driver will be liberated. The entertainment, social, and consumption scenarios during travel may be completely opened up, opening up a trillion-level market.
Autonomous driving currently has two technical routes: Autonomous Driving (AD) and Vehicle-Infrastructure Cooperated Autonomous Driving (VICAD).
AD mainly relies on the vehicle’s own vision, millimeter wave radar, lidar and other sensors, computing units, and wire control systems for environmental perception, calculation decisions, and control execution.
VICAD is based on the intelligent autonomous driving of a single vehicle, through the Internet of Vehicles, organically connects the “people-vehicle-road-cloud” transportation participation elements to help autonomous vehicles in environmental perception, computational decision-making, and control execution. Capability upgrades accelerate the maturity of autonomous driving applications. VICAD can provide safer, more comfortable, more energy-efficient, and more environmentally friendly driving methods. It is also an important link in the city’s intelligent transportation system.
The limitations of bicycle intelligence
First of all, the safety aspect of autonomous driving is the key that affects the commercialization of autonomous driving. In terms of low-level autonomous driving, many car companies have already commercialized mass production. However, many ADAS functions still have the risk of insufficient response capabilities and failure in specific scenarios. For example, for severe weather, tunnel environment, ghost probes, etc., the current automatic driving system cannot perfectly solve these problems. The reliability of automatic driving and the ability to respond to these highly challenging traffic scenarios need improvments.
Secondly, there is the unresolved long-tail problem of bicycle perception. The long-tail problem of perception is one of the reasons currently restricting the design domain of intelligent autonomous driving vehicles. There are more issues: vehicle sensor installation position, detection distance, field of view, data throughput, calibration accuracy, time synchronization, etc… When the vehicle is driving in busy intersections, bad weather, small object perception and recognition, signal light recognition, and other environmental conditions, the vehicle is difficult to completely solve the problems of accurate perception and recognition and high-precision positioning. These long-tail problems have severely restricted and affected the commercialization of autonomous driving on a large scale.
Finally, the economic problems of autonomous driving have not been fully resolved. Economics is a practical issue in the commercialization of autonomous driving on a large scale. In order to achieve high-level autonomous driving, the number of on-board sensors needs to be significantly increased. At present, the hardware devices of L4 autonomous vehicles generally include: 6-12 cameras, 3-12 millimeter wave radars, 5 lidars within 5, and 1～12 cameras. With 2 GNSS/IMU and 1 to 2 computing platforms, the hardware cost is too high and it is difficult to ensure the economy of the vehicle.
The three stages of vehicle-road collaborative autonomous driving
Therefore, we know the current dilemma of bicycle intelligence. And we need to essentially improve the ability of autonomous driving. Vehicle-road collaborative automatic driving can greatly expand the perception range of bicycles and enhance the ability of perception through information interaction coordination, cooperative perception and cooperative decision-making control. And it also introduces new intelligence elements represented by high-dimensional data to realize group intelligence. It can essentially solve the technical bottlenecks encountered by intelligent automatic driving of bicycles, improve the ability of automatic driving, and ensure the safety of automatic driving. VICAD is an advanced development form of AD. It is a gradual development process from low to high. There are three major categories. The stage of development:
(1) Stage 1: Information interaction and coordination, to realize the information interaction and sharing between vehicles and roads;
(2) Phase 2: Cooperative perception. Based on Phase 1, take advantage of roadside perception and positioning, and conduct cooperative perception and positioning with vehicles;
(3) Stage 3: On the basis of stage 1 and stage 2, the vehicle and the road can realize coordinated decision-making and control functions. It can ensure that the vehicle can achieve high-level automatic driving in all road environments.
Vehicles and roads collaborate to solve autonomous driving safety issues
An autonomous vehicle is an extremely complex system. The actual driving environment in which there are many elements and complex changes makes vehicles have many uncertainties. And this uncertainty is more manifested in the perception and prediction level of autonomous driving. For example, in terms of perception, the main difficulties encountered by autonomous driving include extreme weather affecting the reflection effect of lidar, perception errors under adverse lighting conditions, and perception failure under obscured conditions. For prediction, it includes pedestrian and vehicle trajectories. Uncertainty of prediction, reliability of automatic driving decision-making algorithm, etc.
Collaborative perception, collaborative decision-making, and collaborative control are able to greatly improve the safety. Such realization mainly includes two aspects:
1. Convert unsafe scenes to safe scenes
There are two ways to deal with the original “unsafe” scene. One is to improve the ability of automatic driving to transform it into a safe scene. The other is to detect trigger conditions and eliminate them by restricting ODD. The addition of vehicle-road coordination allows autonomous vehicles to obtain more comprehensive data and start processing earlier and farther, thereby creating better conditions for vehicles to respond to unsafe scenarios. At the same time, it also supports the enhancement of trigger detection capabilities for dangerous scenes.
2. Convert an unknown scene to a known scene
The exploration of the original “unknown” scene is an industry problem. “You never know what you don’t know”. On the one hand, vehicle-road collaboration can trigger and process unknown phenomena through full perception and recognition, such as unknown abnormal traffic phenomena. It is transformed into a trigger condition and prompts passing vehicles to make predictions in advance; on the other hand, through data-driven and algorithmic learning, unknown data collection, mining, and training can be improved, and unknown scenarios can be discovered, thereby completing the growth of the learning system.
Therefore, in order to realize the large-scale commercialization of autonomous driving as soon as possible, it is necessary to further carry out in-depth research and development and testing of the vehicle-road collaborative deep integration system. And we have to accelerate the construction and deployment of intelligent roads, and ensure the safe operation of autonomous driving.
