Summary:
The automatic driving system is a comprehensive system that integrates environmental perception, decision-making control, and action execution. It is a system that fully considers the coordinated planning of vehicles and the traffic environment, and is also an important part of the future intelligent transportation system. This paper focuses on the analysis of related technologies for autonomous driving decision-making control, and explores the future development direction.
1. Introduction to autonomous driving system
In general, an autonomous driving system can be divided into a perception layer, a decision-making layer, and an execution layer .
perception layer
The perception layer is defined as the collection and processing of environmental information and in-vehicle information, involving road boundary detection, vehicle detection, pedestrian detection and many other technologies. It can be considered as an advanced sensor technology. The sensors used include lidar, camera , millimeter wave radar, ultrasonic radar, speed and acceleration sensors, etc. Due to the limitation of perception of a single sensor, it cannot meet the precise perception under various working conditions. To achieve smooth operation of autonomous vehicles in various environments, multi-sensor fusion technology is required, which is also the key technology of the perception layer.
decision-making
The decision-making layer can be understood as making decision-making judgments based on perceptual information, determining an appropriate working model, formulating corresponding control strategies, and replacing human drivers to make driving decisions. The function of this part is similar to assigning corresponding tasks to self-driving cars. For example, in systems such as lane keeping, lane departure warning, vehicle distance keeping, and obstacle warning, it is necessary to predict the state of the vehicle and other vehicles, lanes, pedestrians, etc. that it encounters in the future. Advanced decision theory includes fuzzy reasoning, reinforcement learning, neural networks and Bayesian network techniques, etc. Due to the variety of road conditions and scenarios that humans face during driving, and different people have different driving strategies for different situations, the optimization of human-like driving decision-making algorithms requires a very complete and efficient artificial intelligence model and Lots of valid data. These data need to cover various rare road conditions as much as possible, and this is also the biggest bottleneck in the development of driving decisions.
executive layer
The execution layer refers to that the system controls the vehicle according to the decision result after making a decision. Each control system of the vehicle needs to be able to connect with the decision-making system through the bus, and be able to accurately control the driving actions such as acceleration, braking, steering range, and lighting control according to the bus instructions issued by the decision-making system, so as to realize the autonomous driving of the vehicle .
Figure 1 Introduction to the autonomous driving system
2. Introduction to decision-making control system
Traditionally, the decision-making control software system of an automatic driving system includes functional modules such as environment prediction, behavior decision-making, action planning, and path planning .
Environment Prediction Module
As one of the direct data upstream of the decision-making planning control module, the environment prediction module is mainly used to predict the behavior of the objects recognized by the perception layer, and convert the predicted results into trajectories of time and space dimensions and pass them on to subsequent modules. Usually, the object information output by the perception layer includes physical attributes such as position, speed, and direction.
Using these output physical properties, "instantaneous predictions" can be made about objects. The environment prediction module is not limited to making predictions on objects in combination with physical laws, but can combine objects and surrounding environments as well as accumulated historical data information to make more "macroscopic" behavior predictions for perceived objects. For example, in Figure 2, it is predicted that pedestrians may cross the intersection on the sidewalk by recognizing the historical movement of pedestrians on the sidewalk, and it can be judged that they will turn right at the intersection through the historical trajectory of the vehicle.
Figure 2 Schematic diagram of environmental prediction
Behavioral decision-making module
The behavior decision-making module plays the role of "co-pilot" in the entire automatic driving decision-making planning control software system. This level brings together all the important surrounding information of the vehicle, including not only the real-time position, speed, and direction of the self-driving car itself, but also all relevant obstacle information and predicted trajectories within a certain distance around the vehicle. The problem that the behavioral decision-making layer needs to solve is to determine the driving strategy of the self-driving car on the basis of knowing this information.
Due to the need to consider many different types of information, behavioral decision-making problems are often difficult to solve with a single mathematical model, but some advanced concepts of software engineering should be used to design a rule engine system. For example, in the DARPA challenge, Stanford's unmanned vehicle system uses a series of cost designs and finite state machines to design the trajectory and control instructions of the unmanned vehicle. At this stage, the model of the Markov decision process has also begun to be more and more used in the implementation of decision-making algorithms at the behavioral level of the automatic driving system. In short, the behavioral decision-making level needs to combine the results of the environmental prediction module to output macro decision-making instructions for the subsequent planning module to execute more specifically.
Action Planning Module
The autonomous vehicle planning module includes two parts: action planning and path planning. The action planning module is mainly to plan short-term or even instantaneous actions, such as turning, obstacle avoidance, overtaking and other actions; while the path planning module is to plan the vehicle's driving path for a long time, such as from the starting point to the destination. route design or selection.
The design idea of the automatic driving system is to establish several driving states and trigger the switching of driving states through different conditions. This design idea has the problem of poor smoothness in the switching process. In the actual system design process, the method of describing the real and non-real objects on the road as virtual particles is mainly used to enhance the ride comfort of the vehicle. Among them, real targets mainly refer to factors such as vehicles and pedestrians; non-real targets include speed limits, red lights, parking spots, road curvature, weather conditions, etc. The advantage of the method based on the virtual particle model is that the algorithm model is unified, which effectively avoids the problem of vehicle acceleration and deceleration jumps caused by target or control mode switching in traditional control algorithms.
path planning module
The automatic driving vehicle path planning module refers to planning an effective path without collision and safely reaching the target point according to the performance index after the starting point and target point of the automatic driving vehicle are given on the basis of a certain environment model. Path planning mainly includes two steps: establishing an environment map including obstacle areas and free areas, and selecting an appropriate path search algorithm in the environment map to search for a drivable path quickly and in real time. The result of path planning plays a role of navigation for the vehicle, which guides the vehicle from the current position to the target position. Environmental map representation methods are mainly divided into metric map representation, topological map representation and so on.
3. Development trend
The development of technologies such as artificial intelligence machine learning, deep neural network, and networked communication has further enriched the technical path for the development of autonomous driving vehicles, and has also promoted the development of autonomous driving technology from a single prototype demonstration to one with certain application capabilities and self-positioning capabilities. Orientation development of typical traffic scenarios.
artificial intelligence
Artificial intelligence is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence. It intends to explore the essence of intelligence and produce a new kind of intelligent machine that can respond in a similar way to human intelligence. One of its important application fields is automatic driving. The main goal is to enable self-driving cars to have a certain degree of self-learning ability and to form memory cognition for simple traffic environments. At this stage, the main application of artificial intelligence technology in the field of self-driving cars It is reflected in the following aspects.
1. Realize the recognition and cognition of environmental objects
Using multi-eye vision, laser radar, millimeter-wave radar and other sensor devices and recognition algorithms, accurate recognition of multi-curved objects in the actual road environment can be achieved. At the same time, after incorporating deep learning technology, iterative classification can be formed for the three-dimensional space size and feature information of each object, so that the self-driving car has the ability to recognize and recognize various environmental objects.
2. Realize the detection of the driving area
The map collection technology based on advanced sensors can extract detailed road markings (signs, markings, signal lights, etc.) and high-precision location (longitude, latitude, height, etc.) At the same time, based on deep learning, it can realize the recognition and recognition of the drivable and non-drivable areas of the road.
3. Realize the planning and decision-making of the driving route
Decision planning processing is another important application scenario of artificial intelligence technology in autonomous driving. At this stage, mainstream artificial intelligence methods include state machines, decision trees, and Bayesian networks. With the development of deep learning and enhanced learning technology, the decision-making of complex working conditions and online optimization learning have been realized. Since there are many factors that affect driving path planning in actual roads, it will inevitably occupy more computing resources. In order to improve computing efficiency, Japanese researchers put forward the research idea of "safe field", that is, to form typical traffic scenes as the input of deep learning neural network, so as to improve the decision-making efficiency of autonomous vehicles and improve the path planning ability.
Figure 3 Machine learning-based unstructured road detection framework
Intelligent network connection
Combined with the development of communication technology, the use of real-time communication technology between cars and cars, cars and roads, cars and people, and cars and the cloud can provide data, calculations and algorithms for artificial intelligence technology in the application process of autonomous driving technology. It can also provide further support for multi-model and multi-scenario intelligent driving requirements, and provide solutions to the problems faced by the collaborative driving of group intelligent driving systems. The specific architecture of the vehicle-cloud collaborative autonomous driving system based on intelligent network connection is shown in Figure 4 below.
Figure 4 Schematic diagram of the construction scheme of the vehicle-cloud collaborative autonomous driving system based on artificial intelligence
The architecture solution is divided into two parts: the AI-based autonomous driving intelligent terminal and the autonomous driving cloud system based on big data analysis, which together form a vehicle-cloud collaborative integrated automatic system that integrates precise perception of complex environments, intelligent traffic decision-making, and optimized execution of driving control. driving system. In different driving conditions and application scenarios, vehicle-cloud collaboration technology can realize accurate driving environment perception, intelligent traffic decision-making and optimized driving action control, and realize information data interaction and collaboration between the vehicle terminal and the cloud.
The vehicle-cloud collaboration technology for autonomous driving systems based on intelligent networking mainly solves the problems of insufficient fusion of multi-source heterogeneous data and insufficient computing power of front-end equipment, that is, the sampling data of vehicle body sensor nodes (such as GPS/INS data, millimeter-wave radar data) and Multimedia data (such as camera images) are transmitted to the cloud database at a certain frequency, and online processing, offline processing, traceability processing and complex data analysis are performed at the same time. And based on the intelligent driving control model of artificial intelligence integrated application algorithm, it provides a reliable and efficient collaborative control scheme for vehicle decision-making.
The artificial intelligence algorithm application technology cloud platform is the core part of the automatic driving cloud system. It combines machine learning, data mining and other related technologies to analyze perception fusion information and provide decision-making basis for vehicle control planning. And use virtualization technology and network technology to integrate large-scale scalable computing, storage, data, application and other distributed computing resources to complete the learning and training of artificial intelligence model algorithms, realize the training of artificial intelligence models in the cloud, and use the vehicle-cloud collaboration technology to It is deployed on the embedded platform, so that the artificial intelligence algorithm can be deeply applied in the automatic driving system of the car.
At present, the application of network technology in the field of autonomous driving is mainly concentrated in information services and top-level monitoring. Realizing highly automatic driving through the technical route of intelligent network connection still needs to solve difficult problems such as information security, transmission delay, and network coverage before it can be truly applied.
Intelligent Computing Platform
Self-driving cars have gradually transformed from a means of transportation into a new type of mobile intelligent terminal. Changes in vehicle functions and attributes lead to changes in its electrical and electronic architecture, which in turn requires stronger computing, data storage and communication capabilities as a basis. The vehicle-mounted intelligent computing platform is an important solution to meet the above requirements.
The main function of the on-board intelligent computing platform is to use environmental perception data, navigation and positioning information, vehicle real-time data, cloud intelligent computing platform data and other V2X interactive data as input, based on the core of environment perception and positioning, intelligent planning and decision-making, and vehicle motion control. Control algorithm, output driving, transmission, steering and braking and other execution control commands to realize automatic control of the vehicle, and output data to the cloud intelligent computing platform and V2X equipment, and can also realize the man-machine interaction of vehicle driving information through the human-computer interaction interface. interact.
Reprinted from the Internet, the opinions in the article are for sharing and exchange only, and do not represent the position of this official account. If copyright issues are involved, please let us know and we will deal with them in a timely manner.
Source | The forefront of smart driving