An article about autonomous driving decision-making systems

Autonomous driving is a system that integrates perception, decision-making, and control and fully considers the coordinated planning of vehicles and the traffic environment. It is also an important component of future intelligent transportation systems. Just as people need to know the road conditions, understand the direction, avoid obstacles in time, and reach their destination when walking, perception, decision-making, and control are also very important for self-driving cars.

Perception is just like human eyes and ears. By installing hardware equipment such as millimeter wave radar, lidar, and vehicle cameras on self-driving cars, self-driving cars can see To clear the road conditions, the perception system can detect lane lines, vehicles, pedestrians and other traffic participants through perception hardware. The core of its technology is the accuracy of perception hardware detection and the integration of perception results by multiple perception hardware.

Decision-making is like the brain, which can analyze the road conditions and make the next step. Control is to allow the self-driving car to act according to the The brain (decision-making) analyzes the results and takes action. Decision-making can be understood as making judgment decisions based on perceived information, determining an appropriate working model, and formulating corresponding control strategies, which can replace the driver in making further driving decisions.

According to the classification of the Society of Automotive Engineers, autonomous driving is divided into 6 levels from L0 to L5. As the level of autonomous driving continues to improve, autonomous vehicles become more intelligent. Comprehensive, after reaching L5 level, self-driving cars can complete driving actions independently without the need for a driver to operate the car. The improvement of autonomous driving levels mainly reflects the technological improvement of decision-making systems.

Control is like human hands and feet. It controls and executes specific travel actions, allowing self-driving cars to complete the travel process. The control system can complete the turning and acceleration of the vehicle. , lighting control and a series of actions are the final step to realize autonomous driving.

As mentioned above, the decision-making system is like the human brain, which can control the speed, direction and lights of self-driving cars. The smarter the decision-making system, the better the self-driving car. The higher the achievable level. In traditional understanding, the decision-making system covers environmental prediction, action planning, path planning, behavioral decision-making, etc.

If a self-driving car wants to complete the driving process independently, it is not enough to just see the road conditions clearly. It also needs to predict future road conditions. Therefore, the decision-making system needs to complete the environmental analysis. Prediction. Environmental prediction is to predict the traffic environment. Environmental prediction is not limited to making predictions based on physical laws, but can combine objects and surrounding environments as well as accumulated historical data information to make more "macro" predictions of the perceived environment. Behavior prediction, which covers all aspects of traffic participants, such as behavior prediction of perceived vehicles, pedestrians, etc., can determine their next actions, such as speed, location, etc., through the instantaneous actions of vehicles and pedestrians during perception. Direction, etc., can control itself to complete a series of actions such as avoidance, deceleration, and lane change. In addition, environmental prediction also covers the prediction of traffic signals, which includes the understanding and judgment of traffic environments such as speed limit signs, traffic lights, and tidal lanes.

Action planning puts more emphasis on the self-driving car itself. Action planning mainly plans short-term or even instantaneous actions for self-driving. Based on the results of environmental predictions, complete tasks such as turning, Avoidance, overtaking and other actions. At the same time, traffic action planning is also essential, such as speed control on speed-limited sections, parking/driving under traffic lights, lane line selection for tidal lanes, etc., all of which need to be planned in advance. Action planning allows autonomous vehicles to participate in traffic safely and efficiently, making autonomous driving a reality.

In addition to action planning, self-driving cars also need to complete the planning of the driving path, such as path planning from the departure point to the destination. The required path can be selected or designed. This is the decision to automatically How to drive a car through the necessary steps. Path planning can allow the self-driving car to know the road sections it needs to pass through in the general direction, and make timely path adjustments according to passenger needs (shopping, watching movies, etc.), so that the self-driving car can travel on During the travel process, it can not only shorten the time, but also meet the needs of passengers to a great extent, realize route customization, and make the travel process smoother.

Behavioral decision-making falls to the self-driving car itself, through its own real-time location, speed, direction and other information, together with the traffic information obtained in environmental prediction and the path completed in action planning. Referring to planning, etc., the self-driving car can predict possible dangers and upcoming required actions, so that the self-driving car can adjust its own actions.

The autonomous driving decision-making system is a direct reflection of the intelligence of autonomous vehicles and plays a decisive role in the safety of autonomous vehicles. Since it assumes the task of the "brain" of autonomous vehicles, The self-driving decision-making system also determines what level standards the self-driving car can achieve. At this stage, the development of self-driving cars cannot be praised by everyone. The reason is that self-driving cars are not as smart as human drivers and cannot handle extreme situations such as "ghost probes" and "jamming". The ability to make timely adjustments to its own actions under traffic conditions means that self-driving cars are not "intelligent" at this stage.

In the framework of autonomous driving, among the three frameworks of perception, decision-making, and control, the degree of perception depends on the lidar and millimeter waves loaded on the autonomous vehicle. Radars, vehicle cameras and other hardware equipment are just like people who are short-sighted and can wear myopia glasses. If self-driving cars experience inaccurate or untimely perception, they can directly improve the accuracy of perception by replacing the perception hardware with higher technical standards. Spend. The control serves as the execution end. If there is a problem, the hardware of the autonomous vehicle can be replaced to meet the required standards.

The technical improvement of the decision-making system is not achieved through hardware replacement and improvement like the perception system and control system. In order to make self-driving cars as flexible as human-driving cars, The decision-making system requires a lot of deep learning and needs to be able to handle various unexpected road conditions, and this is the most difficult thing.

Whether the decision-making system meets the standards is not like the perception system and the control system, which have clear written regulations and technical standards. In addition to being able to flexibly handle various unexpected road conditions, the decision-making system is far from enough. To be accepted by consumers, the decision-making system needs to be like a "human" and can handle various emergencies based on "human" thinking, and this is the most difficult thing.

The development of decision-making systems will surely become more intelligent with the "feeding" of large amounts of data, making travel safer. With the development of artificial intelligence, deep neural networks and With the development of intelligent network technology, the decision-making system will be further improved. Among them, intelligent network technology can interact with information between cars and cars, cars and people, and cars and traffic, so that self-driving cars can not only "think" but also "communicate", allowing self-driving cars to predict traffic conditions in advance. Making judgments and knowing changes in the traffic environment in advance makes driving more intelligent, which has also led to the further promotion of intelligent network technology.

The autonomous driving decision-making system is the brain of the autonomous vehicle. The improvement of the decision-making system is just like the growth of human beings. The current decision-making system It’s still like a 2 or 3-year-old child who can walk, but will fall and bump inadvertently when walking. When he grows up to a certain age (technology improves), the decision-making system will be able to complete the travel work independently. At this time, it will automatically Driving cars will be able to achieve L5 level, and autonomous driving will also be implemented.

The above is the detailed content of An article about autonomous driving decision-making systems. For more information, please follow other related articles on the PHP Chinese website!