CN109727470B - Complex scene passing decision method for distributed intelligent network-connected automobile intersection - Google Patents
Complex scene passing decision method for distributed intelligent network-connected automobile intersection Download PDFInfo
- Publication number
- CN109727470B CN109727470B CN201910015069.4A CN201910015069A CN109727470B CN 109727470 B CN109727470 B CN 109727470B CN 201910015069 A CN201910015069 A CN 201910015069A CN 109727470 B CN109727470 B CN 109727470B
- Authority
- CN
- China
- Prior art keywords
- vehicle
- intersection
- traffic
- speed
- decision
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 85
- 238000001514 detection method Methods 0.000 claims abstract description 36
- 230000002787 reinforcement Effects 0.000 claims abstract description 25
- 238000004364 calculation method Methods 0.000 claims description 23
- 238000004891 communication Methods 0.000 claims description 14
- 238000013459 approach Methods 0.000 claims description 12
- 238000005516 engineering process Methods 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 5
- 238000010276 construction Methods 0.000 claims description 3
- 238000013209 evaluation strategy Methods 0.000 claims description 2
- 230000004927 fusion Effects 0.000 abstract description 3
- 238000011160 research Methods 0.000 description 6
- 206010039203 Road traffic accident Diseases 0.000 description 4
- 230000006399 behavior Effects 0.000 description 3
- 238000011835 investigation Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000003631 expected effect Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000002620 method output Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
Images
Landscapes
- Traffic Control Systems (AREA)
Abstract
The invention provides a complex scene traffic decision method for a distributed intelligent network-connected automobile intersection, and belongs to the technical field of intelligent traffic. Different from the traditional rule or reinforcement learning-based signal-free intersection decision method, the method introduces a detection mechanism that vehicles violate the traffic rules, and realizes the fusion of the rules and the reinforcement learning method. Firstly, judging whether a vehicle violates a traffic rule at an intersection by adopting a vehicle violation detection algorithm of an implicit curve family; when the vehicles comply with the traffic rules, a rule method is adopted to make a decision on the passing of the vehicles at the intersection; otherwise, the vehicle passing decision at the intersection is made by adopting a reinforcement learning-based method. The method realizes the intelligent networked automobile autonomous traffic decision of the signal lamp-free intersection in the complex traffic scene mixed with the violation of the traffic rules, and provides technical support for improving the traffic safety and the intersection traffic efficiency in the complex environment.
Description
Technical Field
The invention belongs to the technical field of intelligent traffic, and particularly relates to a complex scene traffic decision method for a distributed intelligent network-connected automobile intersection.
Background
Related reports show that casualties caused by road traffic accidents in China are in the top of the world in 2016, and the direct economic loss of the casualties reaches billions. The document (Hangzhou in Ningshu, TCT-based urban road intersection safety evaluation research [ D ]. Changan university, 2016.) records the traffic accident data investigation of a certain representative city in China, and the investigation result shows that the urban intersection is the first high-incidence road section of the traffic accident. The document shows that the violation of traffic rules by pedestrians and automobile drivers is the root cause of high-incidence traffic accidents at intersections by carrying out field investigation on 15 intersections of a certain city.
An Intelligent networked Vehicle (ICV) is a new generation Vehicle equipped with advanced Vehicle-mounted sensors, controllers, actuators, and other devices, and combines V2X communication (Vehicle-to-event, V2X, refers to information exchange between vehicles and the outside world), so as to exchange and share information between vehicles and people, vehicles, roads, and the like, and realize safe, comfortable, energy-saving, and efficient driving, and finally can be operated in place of people. ICV is a necessary trend in the automotive industry. Intersection traffic decisions refer to ICV passing information gathering, processing, and finally making a decision at what speed the vehicle is passing through the intersection. Making a transit decision is the primary task of an ICV to pass through an intersection. Therefore, the research of the ICV passing decision of the complex intersection in the scene of complying with and violating the traffic rules has very important theoretical significance and engineering value.
The main research subject of the intersection traffic decision is the distributed intersection traffic decision without signal control, and the distributed intersection decision based on the rules has already obtained a series of research achievements, mainly comprising a method based on an acceptable gap model, a method based on a dynamic game, a resource lock method based on a conflict table and the like. However, the rule-based distributed intersection decision can achieve the expected effect only when all vehicles obey the traffic rules, and when illegal traffic rule vehicles exist at the intersection, the rule-based distributed intersection decision effect is not ideal.
Reinforcement learning, also known as refitting learning, is an important machine learning method. Reinforcement learning has also been applied to traffic decisions at intersections in recent years. But the intensive learning requires relatively large amount of calculation, and the ICV is difficult to ensure the whole-course online processing.
A certain research result has been obtained on a detection method for a vehicle violating a traffic regulation (traffic lane safety law in the people's republic of china) (see, for example, references 1,2, and 3). At present, the detection method of violating the traffic rules is mainly applied to provide auxiliary reference for the public security traffic administration to identify behaviors violating the traffic rules, but is not directly applied to autonomous decision making of automatic driving vehicles at intersections.
Disclosure of Invention
The invention aims to provide a complex scene traffic decision method for a distributed intelligent network automobile intersection aiming at the requirements.
In order to achieve the purpose, the invention adopts the following technical scheme:
the invention provides a complex scene traffic decision method for a distributed intelligent network automobile intersection, which is characterized by comprising four parts of 1) system platform establishment and parameter setting, 2) vehicle detection and judgment against traffic rules, 3) intersection traffic decision judgment based on a rule method, and 4) intersection traffic decision judgment based on reinforcement learning, wherein the implementation process comprises the following steps:
1) the system platform is built and the parameters are set, and the method specifically comprises the following steps:
1-1) system platform construction:
a vehicle state detection and V2X broadcasting system is built at the intersection, and comprises a road side server, a camera and V2X communication equipment; the cameras are arranged above the incoming direction of each lane of the intersection, so that all the cameras can cover all areas including the intersection; the roadside server is installed at the roadside of the intersection, is connected with each camera and is used for receiving the images collected by all the cameras and processing all data including the images; the V2X communication equipment is arranged on each intelligent networked automobile in the intersection range and used for receiving the intersection map transmitted by the roadside server and the global vehicle state in the intersection range, wherein the global vehicle state comprises license plate numbers, vehicle coordinates, vehicle speeds, vehicle head orientation angles and acquisition time of all vehicles in the intersection range;
1-2) parameter setting:
setting a cycle period T as 1/f, wherein f is the frequency of the intelligent networked automobile receiving the intersection map and the global vehicle state through V2X communication equipment;
setting a right-hand Cartesian rectangular coordinate system of the intersection, wherein the earth coordinate system of the intersection is established by taking the geometric center of the intersection as a zero point, the west to the east as an X axis and the south to the north as a Y axis;
the following four vehicle speeds are set: a) the vehicle speed 1 is a vehicle speed which is selected to be more than 10km/h and less than 20km/h according to the actual traffic condition; b) the speed 2 is the speed which is greater than 20km/h and less than the highest speed limit of the road, and is selected according to the actual traffic condition; c) the speed 3 is a speed command of the vehicle output by the intersection passing decision based on the learning method, and the command is more than or equal to 0km/h and less than or equal to 20 km/h; d) parking, namely the vehicle speed is 0 km/h;
2) the data acquisition and the detection and judgment of the violation of the traffic rules by the vehicle specifically comprise the following steps:
2-1) the clock generating a trigger signal at frequency f starting a new cycle period;
2-2) at the beginning of a new cycle period, the own vehicle receives the intersection map and the global vehicle state at a frequency f through the V2X communication device;
2-3) taking the intersection map obtained in the step 2-2) and the global vehicle state as input, taking whether each vehicle in the intersection range complies with the traffic rules as output, and judging whether any vehicle in the intersection range violates the traffic rules according to a violation traffic rule detection algorithm; when all vehicles within the range of the intersection comply with the traffic rules, executing the step 3-1); when at least one vehicle in the range of the intersection violates the traffic rules, executing the step 4-1); the detection algorithm for violating the traffic rules comprises a vehicle violation detection algorithm of an implicit curve family, a vehicle violation detection method based on a computer vision technology and a traffic violation detection method based on a vehicle track;
3) the intersection passing decision making judgment based on the rule method specifically comprises the following steps:
3-1) judging whether the self vehicle is in the range of the intersection or not through the coordinates of the self vehicle in the intersection geodetic coordinate system and an intersection map: when the coordinates of the vehicle are within the range of the intersection, outputting a vehicle speed command speed 1 of the vehicle, finishing the calculation of the cycle period, driving the intelligent networked vehicle according to the command speed 1, returning to the step 2-1), and waiting for a clock to generate a trigger signal to start a new cycle period; when the coordinates of the self vehicle are not in the range of the intersection, executing the step 3-2);
3-2) under an intersection geodetic coordinate system, defining threshold time as the vertical distance between the own vehicle coordinate and an intersection stop line divided by the own vehicle speed, and defining the condition that the own vehicle coordinate is close to the intersection when the threshold time is less than 2/f, otherwise, defining that the own vehicle coordinate is not close to the intersection; judging whether the coordinates of the vehicle are close to the intersection: when the coordinates of the vehicle do not approach the intersection, outputting a vehicle speed command speed 2 of the vehicle, finishing the calculation of the cycle period, driving the intelligent networked vehicle according to the command speed 2, returning to the step 2-1), and waiting for a clock to generate a trigger signal to start a new cycle period; when the coordinates of the self vehicle approach to the intersection, executing the step 3-3);
3-3) carrying out intersection passing decision on the self-vehicle by adopting an intersection decision based on a rule method according to the intersection map and the global vehicle state received in the step 2-2); then outputting a speed instruction of the vehicle as a speed 1 or stopping according to a rule method-based intersection decision result, finishing the calculation of the cycle period, enabling the intelligent networked vehicle to run or stop according to the speed 1 instruction, returning to the step 2-1), and waiting for a clock to generate a trigger signal to start a new cycle period; the intersection decision based on the rule method comprises an acceptable gap model-based method, a dynamic game-based control method, a conflict table-based resource lock method and a vehicle passing rule base-based method;
4) the intersection passing decision based on reinforcement learning specifically comprises the following steps:
4-1) judging whether the coordinates of the own vehicle are in the intersection range according to the intersection map and the global vehicle state received in the step 2-2): when the coordinates of the self vehicle are not in the range of the intersection, executing the step 4-2); when the coordinates of the self-vehicle are within the range of the intersection, executing the step 4-3);
4-2) judging whether the coordinates of the self-vehicle are close to the intersection by adopting whether the coordinates of the self-vehicle defined in the step 3-2) are close to the intersection: when the coordinates of the vehicle do not approach the intersection, outputting a vehicle speed command speed 2 of the vehicle, finishing the calculation of the cycle period, driving the intelligent networked vehicle according to the command speed 2, returning to the step 2-1), and waiting for a clock to generate a trigger signal to start a new cycle period; when the coordinates of the vehicle approach to the intersection, outputting a vehicle speed instruction of the vehicle to stop, namely the vehicle speed is 0km/s, finishing the calculation of the cycle period, stopping the intelligent network-connected vehicle, returning to the step 2-1), and waiting for a clock to generate a trigger signal to start a new cycle period;
4-3) carrying out intersection passing decision on the self-vehicle by adopting reinforcement learning based on a time sequence difference method according to the intersection map and the global vehicle state received in the step 2-2); then, according to a cross decision result obtained by reinforcement learning based on a time sequence difference method, outputting a speed instruction of the vehicle as a vehicle speed 3, finishing the calculation of the cycle period, enabling the intelligent networked vehicle 4 to run according to the instruction vehicle speed 3, returning to the step 2-1), and waiting for a clock to generate a trigger signal to start a new cycle period; the reinforcement learning based on the time sequence difference method comprises a Sarsa algorithm with the same strategy and a Q-learning algorithm with a different strategy.
The invention has the characteristics and beneficial effects that: different from the traditional rule-based distributed intersection decision method or the traditional reinforcement learning-based intersection decision method, the invention introduces a detection mechanism that vehicles violate traffic rules into the autonomous decision of the automatically-driven vehicles at the intersection aiming at the intersection without signal lamps under the condition of full-automatic driving, provides the distributed intelligent network-connected automobile intersection complex scene traffic decision method, and realizes the fusion of the rules and the learning decision method. Judging whether vehicles at the intersection violate the traffic rules by adopting a vehicle violation traffic rule detection mechanism; when the vehicles obey the traffic rules, a rule method is adopted to make the vehicle passing decision at the intersection; and when vehicles violating the traffic rules exist at the intersection, a learning method is adopted to make the decision of vehicle passing at the intersection. The method realizes the intelligent networked automobile autonomous traffic decision of the signal lamp-free intersection in the complex traffic scene mixed with the violation of the traffic rules, and provides technical support for improving the traffic safety and the intersection traffic efficiency in the complex environment.
Drawings
Fig. 1 is a schematic flow chart of a complex scene traffic decision method for a distributed intelligent network-connected automobile intersection according to an embodiment of the present invention.
Detailed Description
The technical scheme of the invention is described in detail below with reference to the accompanying drawings and embodiments:
the invention provides a complex scene traffic decision method for a distributed intelligent network-connected automobile intersection, which is shown in a flow chart of fig. 1 and comprises four parts of 1) system platform establishment and parameter setting, 2) vehicle violation traffic rule detection and judgment, 3) intersection traffic decision judgment based on a rule method and 4) intersection traffic decision judgment based on reinforcement learning, and the specific implementation process is as follows:
1) the system platform is built and the parameters are set, and the method specifically comprises the following steps:
1-1) system platform construction:
a vehicle state detection and V2X broadcasting system is built at the intersection, and comprises a road side server, a camera and V2X communication equipment; the cameras are arranged above the incoming direction of each lane of the intersection, so that all the cameras can cover all areas including the intersection; the roadside server is installed at the roadside of the intersection, is connected with each camera and is used for receiving the images collected by all the cameras and processing all data including the images; the V2X communication equipment is arranged on each intelligent networked automobile in the intersection range and used for receiving the intersection map transmitted by the roadside server and the global vehicle state in the intersection range, wherein the global vehicle state comprises license plate numbers, vehicle coordinates, vehicle speeds, vehicle head orientation angles and acquisition time of all vehicles in the intersection range;
the built system platform is used for detecting information such as global vehicle states (namely license plate numbers, vehicle coordinates, vehicle speeds, vehicle head orientation angles and acquisition time of all vehicles) in an intersection range in real time, communicating with the ICV through the V2X communication facility 5, broadcasting an intersection map (namely coordinates of roads on which vehicles can run at the intersection) and the global vehicle states to each ICV in the intersection range, and each ICV in the intersection range is provided with a vehicle-mounted V2X device and used for receiving the information such as the license plate numbers, the vehicle coordinates, the vehicle speeds, the vehicle head orientation angles and the acquisition time of all vehicles in the intersection range and the intersection map.
1-2) parameter setting:
setting a cycle period T as 1/f, wherein f is the frequency of receiving the intersection map and the global vehicle state by the ICV through the vehicle-mounted V2X communication equipment, and the value of the cycle period T is selected according to the actual situation and generally ranges from 10ms to 200 ms.
The earth coordinate system of the intersection is set as a right-hand Cartesian rectangular coordinate system which is established by taking the geometric center of the intersection as a zero point, the west to the east as an X axis and the south to the north as a Y axis.
The following four vehicle speeds are set: a) the vehicle speed 1 is a vehicle speed which is selected to be more than 10km/h and less than 20km/h according to the actual traffic condition; b) the speed 2 is the speed which is greater than 20km/h and less than the highest speed limit of the road, and is selected according to the actual traffic condition; c) the speed 3 is a speed command of the vehicle output by the intersection passing decision judgment based on the learning method, and the command is more than or equal to 0km/h and less than or equal to 20 km/h; d) parking, i.e. a vehicle speed of 0 km/h.
2) The data acquisition and the detection and judgment of the violation of the traffic rules by the vehicle specifically comprise the following steps:
2-1) clock generates a trigger signal at frequency f to start a new cycle.
2-2) at the beginning of a new cycle period, the own vehicle receives the intersection map and the global vehicle state at a frequency f through the V2X communication device;
2-3) taking the intersection map obtained in the step 2-2) and the global vehicle state as input, taking whether each vehicle in the intersection range complies with the traffic rules as output, and judging whether any vehicle in the intersection range violates the traffic rules (namely violates the traffic road safety law of the people's republic of China) according to the detection algorithm violating the traffic rules; when all vehicles within the range of the intersection comply with the traffic rules, executing the step 3-1); when at least one vehicle in the range of the intersection violates the traffic rules, executing the step 4-1); the detection algorithm for violating the traffic rules mainly comprises a vehicle violation detection algorithm of an implicit curve family (specifically, refer to reference 1), a vehicle violation detection method based on a computer vision technology (specifically, refer to reference 2), a traffic violation detection method based on a vehicle track (specifically, refer to reference 3) and the like;
the embodiment adopts a vehicle violation detection algorithm of an implicit curve family, and the specific implementation mode is as follows:
2-3-1) describing different areas on the intersection map by adopting an implicit curve family F according to the actual situation of the intersection on a ground coordinate system of the intersection, wherein F is { F ═ Fi(x, y) }, i ═ 1,2, …, n, i denotes the i-th area on the intersection map, (x, y) denotes the vehicle position, and the implicit curve f denotes the position of the vehiclei(x,y)>0、fi(x,y)＝0、fi(x,y)<0 represents that the vehicle is in the ith implicit curve fi (x, y) and the vehicle is in the ith implicit curve fiThe vehicle on the (x, y) th implicit curve fi(x, y) is not;
2-3-2) setting all rule (j) of violation rules according to the traffic road safety law of the people's republic of China, wherein j represents the jth vehicle violation condition specified by the traffic road safety law of the people's republic of China; the initial value of the implicit curve in the 1 st frame is set to 0, i.e., F ═ Fi(x, y) ═ 0}, i ═ 1,2, …, n; then according to the intersection map and the global vehicle state received in the step 2-2),calculating the value of the implicit curve family F of the current cycle period, and judging whether the values of the implicit curve family F of the previous cycle period and the current cycle period meet rule (j); if the rule (j) is satisfied, the corresponding traffic rule is violated, otherwise, the corresponding traffic rule is not violated;
the straight-going left-turn violation is taken as an example for explanation: setting an implicit curve f1(x, y) indicates that only straight lines are allowed in the region 1, and an implicit curve f is set2(x, y) indicates region 2 that the vehicle entered if left turning from region 1, setting the straight left turn violation to Rule (1):
f1(x,y)>＝0→f2(x,y)>when the loop cycle jumps to the current loop cycle from the last loop cycle, the value of the implicit curve changes; then judging whether the vehicles in the previous cycle period and the current cycle period meet the Rule (1) of the left turn Rule of straight going of the vehicles according to the intersection map and the global vehicle state received in the step 2-2); if the right-hand-left-turning rule is satisfied, the vehicle is in a straight-going left-turning rule violation, otherwise, the vehicle is not in a straight-going left-turning rule violation;
3) the intersection passing decision making judgment based on the rule method specifically comprises the following steps:
3-1) judging whether the self vehicle is in the range of the intersection or not through the coordinates of the self vehicle in the intersection geodetic coordinate system and an intersection map: when the coordinates of the vehicle are within the range of the intersection, outputting a vehicle speed command speed 1 of the vehicle, finishing the calculation of the cycle period, then driving the intelligent networked vehicle according to the command speed 1, returning to the step 2-1), and waiting for a clock to generate a trigger signal to start a new cycle period; when the coordinates of the self vehicle are not in the range of the intersection, executing the step 3-2);
3-2) under the intersection geodetic coordinate system, defining the threshold time as the vertical distance between the own vehicle coordinate and the intersection stop line divided by the own vehicle speed, and defining the condition that the own vehicle coordinate is close to the intersection when the threshold time is less than 2/f, otherwise, the own vehicle coordinate is not close to the intersection. And judging whether the coordinates of the vehicle are close to the intersection. When the coordinates of the vehicle do not approach the intersection, outputting a vehicle speed command speed 2 of the vehicle, finishing the calculation of the cycle period, then enabling the intelligent networked vehicle to run according to the command speed 2, returning to the step 2-1), and waiting for a clock to generate a trigger signal to start a new cycle period; when the coordinates of the self vehicle approach to the intersection, executing the step 3-3);
3-3) carrying out intersection passing decision on the self-vehicle by adopting an intersection decision based on a rule method according to the intersection map and the global vehicle state received in the step 2-2); then outputting a speed instruction of the vehicle as a vehicle speed 1 according to a rule method-based intersection decision result, finishing the calculation of the cycle period, then enabling the intelligent networked vehicle to run or stop according to the instruction speed 1, returning to the step 2-1), and waiting for a clock to generate a trigger signal to start a new cycle period; the intersection decision based on the rule method mainly comprises a method based on an acceptable gap model, a control method based on a dynamic game, a resource lock method based on a conflict table, a vehicle passing rule base method and the like;
the intersection decision based on the rule method in the embodiment adopts a rule base method based on vehicle passing, and the specific implementation mode is as follows: setting the priority of the ambulance and the fire truck to be the highest, and setting the priorities of the other vehicles to be the same and lower than the priorities of the ambulance and the fire truck; setting a vehicle speed command after the vehicle decelerates as a speed 1; then, a vehicle passing rule base-based method is adopted for calculation (see the reference 4, page 129 and fig. 6.9);
4) the intersection passing decision based on reinforcement learning specifically comprises the following steps:
4-1) judging whether the coordinates of the own vehicle are within the range of the intersection according to the intersection map and the global vehicle state received in the step 2-2). When the coordinates of the self vehicle are not in the range of the intersection, executing the step 4-2); when the coordinates of the self-vehicle are within the range of the intersection, executing the step 4-3);
4-2) judging whether the coordinates of the self-vehicle are close to the intersection by adopting whether the coordinates of the self-vehicle defined in the step 3-3) are close to the intersection. When the coordinates of the vehicle do not approach the intersection, outputting a vehicle speed command speed 2 of the vehicle, finishing the calculation of the cycle period, then driving the intelligent networked vehicle according to the command speed 2, returning to the step 2-1), and waiting for a clock to generate a trigger signal to start a new cycle period; when the coordinates of the vehicle approach to the intersection, outputting a vehicle speed instruction of the vehicle to stop, namely the vehicle speed is 0km/s, finishing the calculation of the cycle period, stopping the intelligent network-connected vehicle, returning to the step 2-1), and waiting for a clock to generate a trigger signal to start a new cycle period;
4-3) carrying out intersection passing decision on the self-vehicle by adopting reinforcement learning based on a time sequence difference method according to the intersection map and the global vehicle state received in the step 2-2); then, according to a cross decision result obtained by reinforcement learning based on a time sequence difference method, outputting a speed instruction of the vehicle as a vehicle speed 3, finishing the calculation of the cycle period, enabling the intelligent networked vehicle to run according to the instruction vehicle speed 3, returning to the step 2-1), and waiting for a clock to generate a trigger signal to start a new cycle period; the reinforcement learning based on the time sequence difference method comprises a Sarsa algorithm with the same strategy and a Q-learning algorithm with different strategies;
the Sarsa algorithm with the same strategy is adopted in the embodiment, and the specific implementation mode is as follows:
taking a meter as a distance interval, and a is more than or equal to 0.5m and less than or equal to 2m, respectively drawing straight lines along an X axis and a Y axis of a geodetic coordinate system parallel to the intersection, dividing the received intersection map into a plurality of grids, and taking the set of all grids in the intersection range as a state set S; selecting a set consisting of a plurality of speeds as an action set A between 0km/h and 20km/h by using bkm/h as a speed interval, wherein b is more than or equal to 0km/h and less than or equal to 10 km/h; and defining the set track as a group of track points which accord with the traffic road safety law of the people's republic of China and enable the intelligent networked automobile to pass through the intersection. The return of the self vehicle along the set track is r1, and r1 is more than 0 and less than 100; the return of the deviation of the self vehicle from the set track is r2, -50 < r2 < 0; setting the return of collision between the vehicle and other vehicles as r3, -100 < r3 < -50; setting an attenuation factor gamma, wherein gamma is more than 0 and less than 1; setting the evaluation strategy and the action strategy as greedy strategies;
then, the calculation is performed according to the Sarsa algorithm of the same strategy (see reference 5, page 79, fig. 5.6 for the Sarsa algorithm of the same strategy), and the calculation result is set as the vehicle speed command of the output vehicle, namely, the speed 3.
In one period, the method outputs at most one vehicle speed command, namely one command of parking, vehicle speed 1, vehicle speed 2 and vehicle speed 3; the ICV then responds to the vehicle speed command and drives the ICV through the intersection at the vehicle speed given by the command until the ICV leaves the intersection area.
Different from the traditional rule-based distributed intersection decision method or the reinforcement learning-based intersection decision method, the invention provides the distributed intelligent network-connected automobile intersection traffic decision method by introducing a detection mechanism that vehicles violate traffic rules into the autonomous decision of the automatically-driven vehicles of the intersection aiming at the intersection without signal lamps under the condition of full-automatic driving, thereby realizing the fusion of the rules and the learning decision method. The method realizes the intelligent networked automobile autonomous traffic decision of the signal lamp-free intersection in the complex traffic scene mixed with the violation of the traffic rules, and provides technical support for improving the traffic safety and the intersection traffic efficiency in the complex environment.
The above-mentioned embodiments are merely illustrative of the preferred embodiments of the present invention, and do not limit the scope of the present invention, and variations and modifications of the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.
The literature references describe:
reference 1: marliwen, Guo YuKun, Li jin Screen, a vehicle violation detection algorithm [ J ] by implicit Curve family, university of electronic technology, Western Ann university of science (Nature science edition), 2016,43(2): 139-.
Reference 2: violation identification algorithm research in intelligent transportation [ D ]. harbourabin industry university, 2014.
Reference 3: jia Yonghua, Zhang Xiao, Zhujiang, a traffic violation behavior detection method based on vehicle track [ J ] China public safety, 2012(11): 196-.
Reference 4: luguangquan, wangynpeng, new field vehicle cooperative safety control technology [ M ]. scientific press, 2014. (page 129, fig. 6.9 no-signal intersection driving behavior flow chart).
Reference 5: guo constitution, Youguan purity, deep and shallow reinforcement learning [ M ], electronic industry Press, 2018 (page 79, FIG. 5.6, same strategy as Sarsa reinforcement learning algorithm).
Claims (5)
1. A complex scene traffic decision method for a distributed intelligent network automobile intersection comprises 1) system platform building and parameter setting, 2) vehicle detection and judgment against traffic rules, 3) intersection traffic decision judgment based on a rule method, and 4) intersection traffic decision judgment based on reinforcement learning, and is specifically realized in the following steps:
1) the system platform is built and the parameters are set, and the method specifically comprises the following steps:
1-1) system platform construction:
a vehicle state detection and V2X broadcasting system is built at the intersection, and comprises a road side server, a camera and V2X communication equipment; the cameras are arranged above the incoming direction of each lane of the intersection, so that all the cameras can cover all areas including the intersection; the roadside server is installed at the roadside of the intersection, is connected with each camera and is used for receiving the images collected by all the cameras and processing all data including the images; the V2X communication equipment is arranged on each intelligent networked automobile in the intersection range and used for receiving the intersection map transmitted by the roadside server and the global vehicle state in the intersection range, wherein the global vehicle state comprises license plate numbers, vehicle coordinates, vehicle speeds, vehicle head orientation angles and acquisition time of all vehicles in the intersection range;
its characterized in that, 1) system platform is built and parameter setting still includes:
1-2) parameter setting:
setting a cycle period T as 1/f, wherein f is the frequency of the intelligent networked automobile receiving the intersection map and the global vehicle state through V2X communication equipment;
setting a right-hand Cartesian rectangular coordinate system of the intersection, wherein the earth coordinate system of the intersection is established by taking the geometric center of the intersection as a zero point, the west to the east as an X axis and the south to the north as a Y axis;
the following four vehicle speeds are set: a) the vehicle speed 1 is a vehicle speed which is selected to be more than 10km/h and less than 20km/h according to the actual traffic condition; b) the speed 2 is the speed which is greater than 20km/h and less than the highest speed limit of the road, and is selected according to the actual traffic condition; c) the speed 3 is a speed command of the vehicle output by the intersection passing decision based on the learning method, and the command is more than or equal to 0km/h and less than or equal to 20 km/h; d) parking, namely the vehicle speed is 0 km/h;
2) the data acquisition and the detection and judgment of the violation of the traffic rules by the vehicle specifically comprise the following steps:
2-1) the clock generating a trigger signal at frequency f starting a new cycle period;
2-2) at the beginning of a new cycle period, the own vehicle receives the intersection map and the global vehicle state at a frequency f through the V2X communication device;
2-3) taking the intersection map obtained in the step 2-2) and the global vehicle state as input, taking whether each vehicle in the intersection range complies with the traffic rules as output, and judging whether any vehicle in the intersection range violates the traffic rules according to a violation traffic rule detection algorithm; when all vehicles within the range of the intersection comply with the traffic rules, executing the step 3-1); when at least one vehicle in the range of the intersection violates the traffic rules, executing the step 4-1); the detection algorithm for violating the traffic rules comprises a vehicle violation detection algorithm of an implicit curve family, a vehicle violation detection method based on a computer vision technology and a traffic violation detection method based on a vehicle track;
3) the intersection passing decision making judgment based on the rule method specifically comprises the following steps:
3-1) judging whether the self vehicle is in the range of the intersection or not through the coordinates of the self vehicle in the intersection geodetic coordinate system and an intersection map: when the coordinates of the vehicle are within the range of the intersection, outputting a vehicle speed command speed 1 of the vehicle, finishing the calculation of the cycle period, driving the intelligent networked vehicle according to the command speed 1, returning to the step 2-1), and waiting for a clock to generate a trigger signal to start a new cycle period; when the coordinates of the self vehicle are not in the range of the intersection, executing the step 3-2);
3-2) under an intersection geodetic coordinate system, defining threshold time as the vertical distance between the own vehicle coordinate and an intersection stop line divided by the own vehicle speed, and defining the condition that the own vehicle coordinate is close to the intersection when the threshold time is less than 2/f, otherwise, defining that the own vehicle coordinate is not close to the intersection; judging whether the coordinates of the vehicle are close to the intersection: when the coordinates of the vehicle do not approach the intersection, outputting a vehicle speed command speed 2 of the vehicle, finishing the calculation of the cycle period, driving the intelligent networked vehicle according to the command speed 2, returning to the step 2-1), and waiting for a clock to generate a trigger signal to start a new cycle period; when the coordinates of the self vehicle approach to the intersection, executing the step 3-3);
3-3) carrying out intersection passing decision on the self-vehicle by adopting an intersection decision based on a rule method according to the intersection map and the global vehicle state received in the step 2-2); then outputting a speed instruction of the vehicle as a speed 1 or stopping according to a rule method-based intersection decision result, finishing the calculation of the cycle period, enabling the intelligent networked vehicle to run or stop according to the speed 1 instruction, returning to the step 2-1), and waiting for a clock to generate a trigger signal to start a new cycle period; the intersection decision based on the rule method comprises an acceptable gap model-based method, a dynamic game-based control method, a conflict table-based resource lock method and a vehicle passing rule base-based method;
4) the intersection passing decision based on reinforcement learning specifically comprises the following steps:
4-1) judging whether the coordinates of the own vehicle are in the intersection range according to the intersection map and the global vehicle state received in the step 2-2): when the coordinates of the self vehicle are not in the range of the intersection, executing the step 4-2); when the coordinates of the self-vehicle are within the range of the intersection, executing the step 4-3);
4-2) judging whether the coordinates of the self-vehicle are close to the intersection by adopting whether the coordinates of the self-vehicle defined in the step 3-2) are close to the intersection: when the coordinates of the vehicle do not approach the intersection, outputting a vehicle speed command speed 2 of the vehicle, finishing the calculation of the cycle period, driving the intelligent networked vehicle according to the command speed 2, returning to the step 2-1), and waiting for a clock to generate a trigger signal to start a new cycle period; when the coordinates of the vehicle approach to the intersection, outputting a vehicle speed instruction of the vehicle to stop, namely the vehicle speed is 0km/s, finishing the calculation of the cycle period, stopping the intelligent network-connected vehicle, returning to the step 2-1), and waiting for a clock to generate a trigger signal to start a new cycle period;
4-3) carrying out intersection passing decision on the self-vehicle by adopting reinforcement learning based on a time sequence difference method according to the intersection map and the global vehicle state received in the step 2-2); then, according to a cross decision result obtained by reinforcement learning based on a time sequence difference method, outputting a speed instruction of the vehicle as a vehicle speed 3, finishing the calculation of the cycle period, enabling the intelligent networked vehicle to run according to the instruction vehicle speed 3, returning to the step 2-1), and waiting for a clock to generate a trigger signal to start a new cycle period; the reinforcement learning based on the time sequence difference method comprises a Sarsa algorithm with the same strategy and a Q-learning algorithm with a different strategy.
2. The distributed intelligent internet automobile intersection complex scene traffic decision method as claimed in claim 1, wherein in step 2-3), the detection algorithm violating the traffic rules adopts a vehicle violation detection algorithm of an implicit curve family, and the specific implementation process is as follows:
2-3-1) describing different areas on the intersection map by adopting an implicit curve family F according to the actual situation of the intersection on a ground coordinate system of the intersection, wherein F is { F ═ Fi(x, y) }, i ═ 1,2, …, n, i denotes the i-th area on the intersection map, (x, y) denotes the vehicle position, and the implicit curve f denotes the position of the vehiclei(x,y)>0、fi(x,y)＝0、fi(x,y)<0 respectively represents the vehicle in the ith implicit curve fiImplicit curve f of vehicle in the ith (x, y)iThe vehicle on the (x, y) th implicit curve fi(x, y) is not;
2-3-2) setting all rule (j) of violation rules according to the traffic road safety law of the people's republic of China, wherein j represents the jth vehicle violation condition specified by the traffic road safety law of the people's republic of China; the initial value of the implicit curve in the 1 st frame is set to 0, i.e., F ═ Fi(x,y) 0, i is 1,2, …, n; then, calculating the value of an implicit curve family F of the current cycle period according to the intersection map and the global vehicle state received in the step 2-2), and judging whether the values of the implicit curve family F of the previous cycle period and the current cycle period meet rule (j); if rule (j) is satisfied, the corresponding traffic rule is violated, otherwise, the corresponding traffic rule is not violated.
3. The distributed intelligent networked automobile intersection complex scene traffic decision method as claimed in claim 1, wherein in step 3-3), the intersection decision based on the rule method adopts a vehicle traffic rule base method, wherein the priority of the ambulance and the fire truck is set to be the highest, the priority of the other vehicles is the same and lower than that of the ambulance and the fire truck, and the vehicle speed command after the vehicle is decelerated is set to be the speed 1.
4. The distributed intelligent networked automobile intersection complex scene traffic decision method as claimed in claim 1, wherein in step 4-3), the reinforcement learning based on the time sequence difference method adopts a Sarsa algorithm with the same strategy, and the specific implementation manner is as follows:
taking a meter as a distance interval, and a is more than or equal to 0.5m and less than or equal to 2m, respectively drawing straight lines along an X axis and a Y axis of a geodetic coordinate system parallel to the intersection, dividing the received intersection map into a plurality of grids, and taking the set of all grids in the intersection range as a state set S; selecting a set consisting of a plurality of speeds as an action set A between 0km/h and 20km/h by using bkm/h as a speed interval, wherein b is more than or equal to 0km/h and less than or equal to 10 km/h; defining a set track as a group of track points which accord with the traffic road safety law of the people's republic of China and enable the intelligent networked automobile to pass through the intersection; the return of the self vehicle along the set track is r1, and r1 is more than 10 and less than 100; the return of the deviation of the self vehicle from the set track is r2, -50 < r2 < 0; setting the return of collision between the vehicle and other vehicles as r3, -100 < r3 < -50; setting an attenuation factor gamma, wherein gamma is more than 0 and less than 1; setting the evaluation strategy and the action strategy as greedy strategies;
and then, operation is carried out according to the Sarsa algorithm of the same strategy, and the calculation result is set as the vehicle speed 3.
5. The distributed intelligent network-connected automobile intersection complex scene traffic decision method according to any one of claims 1 to 4, characterized in that in step 1-2), the set cycle period T is in a range of 10ms to 200 ms.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910015069.4A CN109727470B (en) | 2019-01-08 | 2019-01-08 | Complex scene passing decision method for distributed intelligent network-connected automobile intersection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910015069.4A CN109727470B (en) | 2019-01-08 | 2019-01-08 | Complex scene passing decision method for distributed intelligent network-connected automobile intersection |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109727470A CN109727470A (en) | 2019-05-07 |
| CN109727470B true CN109727470B (en) | 2020-09-11 |
Family
ID=66298875
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910015069.4A Active CN109727470B (en) | 2019-01-08 | 2019-01-08 | Complex scene passing decision method for distributed intelligent network-connected automobile intersection |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109727470B (en) |
Families Citing this family (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110111605B (en) * | 2019-06-12 | 2021-08-31 | 吉林大学 | On-ramp driving decision-making method for autonomous vehicles based on dynamic game |
| CN110263709B (en) * | 2019-06-19 | 2021-07-16 | 百度在线网络技术（北京）有限公司 | Driving decision mining method and device |
| CN110288847B (en) * | 2019-06-28 | 2021-01-19 | 浙江吉利控股集团有限公司 | An automatic driving decision-making method, device, system, storage medium and terminal |
| CN110444015B (en) * | 2019-07-08 | 2020-10-09 | 清华大学 | Speed decision-making method of intelligent networked vehicles based on zoning of unsignaled intersections |
| CN110473419A (en) * | 2019-09-09 | 2019-11-19 | 重庆长安汽车股份有限公司 | A kind of passing method of automatic driving vehicle in no signal lamp intersection |
| CN113205693A (en) * | 2020-01-31 | 2021-08-03 | 奥迪股份公司 | Method for operating a traffic signal system |
| CN111833614A (en) * | 2020-09-18 | 2020-10-27 | 得威科技(浙江)有限公司 | Traffic violation automatic identification method, system, electronic equipment and storage medium |
| CN112349094B (en) * | 2020-09-27 | 2021-10-22 | 北京博研智通科技有限公司 | A method and system for evaluating the traffic efficiency of motor vehicles at intersections without signal lights |
| CN112373472B (en) * | 2021-01-14 | 2021-04-20 | 长沙理工大学 | Method for controlling vehicle entering time and running track at automatic driving intersection |
| CN113276884B (en) * | 2021-04-28 | 2022-04-26 | 吉林大学 | Intelligent vehicle interactive decision passing method and system with variable game mode |
| CN113313957B (en) * | 2021-05-30 | 2022-07-05 | 南京林业大学 | Vehicle scheduling method at intersection without signal lights based on enhanced Dijkstra algorithm |
| CN114268369A (en) * | 2021-12-20 | 2022-04-01 | 东南大学 | An Inter-Vehicle Communication System Based on Visible Light |
| CN114919581B (en) * | 2022-05-11 | 2024-04-26 | 中南大学 | Behavior decision method and computer device for intelligent vehicle at disordered intersection |
| CN116168550B (en) * | 2022-12-30 | 2024-07-26 | 福州大学 | Traffic coordination method for intelligent network-connected vehicles at signalless intersections |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE3405269C2 (en) * | 1984-02-15 | 1986-09-25 | Battelle-Institut E.V., 6000 Frankfurt | Method and device for the automatic guidance of vehicles along guidelines |
| CN103177596A (en) * | 2013-02-25 | 2013-06-26 | 中国科学院自动化研究所 | Automatic intersection management and control system |
| CN104590274A (en) * | 2014-11-26 | 2015-05-06 | 浙江吉利汽车研究院有限公司 | Driving behavior self-adaptation system and method |
| CN104754011A (en) * | 2013-12-31 | 2015-07-01 | 中国移动通信集团公司 | Internet of Vehicles multi-party coordination control method and Internet of Vehicles coordination platform |
| CN104882008A (en) * | 2015-06-03 | 2015-09-02 | 东南大学 | Method for vehicle cooperative control at non-signaled intersection in vehicle networking environment |
| JP5973447B2 (en) * | 2010-10-05 | 2016-08-23 | グーグル インコーポレイテッド | Zone driving |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20130007754A (en) * | 2011-07-11 | 2013-01-21 | 한국전자통신연구원 | Apparatus and method for controlling vehicle at autonomous intersection |
| CN103996312B (en) * | 2014-05-23 | 2015-12-09 | 北京理工大学 | Autonomous Vehicle Control System with Social Behavioral Interaction |
| US9459623B1 (en) * | 2015-04-29 | 2016-10-04 | Volkswagen Ag | Stop sign intersection decision system |
| CN105321362B (en) * | 2015-10-30 | 2017-10-13 | 湖南大学 | A kind of intelligent coordinated passing method of intersection vehicles |
| US9983591B2 (en) * | 2015-11-05 | 2018-05-29 | Ford Global Technologies, Llc | Autonomous driving at intersections based on perception data |
| CN108009587B (en) * | 2017-12-01 | 2021-04-16 | 驭势科技（北京）有限公司 | Method and equipment for determining driving strategy based on reinforcement learning and rules |
| CN108225364B (en) * | 2018-01-04 | 2021-07-06 | 吉林大学 | A system and method for driving task decision-making of an unmanned vehicle |
| CN108877269B (en) * | 2018-08-20 | 2020-10-27 | 清华大学 | A method for vehicle status detection and V2X broadcasting at intersections |
-
2019
- 2019-01-08 CN CN201910015069.4A patent/CN109727470B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE3405269C2 (en) * | 1984-02-15 | 1986-09-25 | Battelle-Institut E.V., 6000 Frankfurt | Method and device for the automatic guidance of vehicles along guidelines |
| JP5973447B2 (en) * | 2010-10-05 | 2016-08-23 | グーグル インコーポレイテッド | Zone driving |
| CN103177596A (en) * | 2013-02-25 | 2013-06-26 | 中国科学院自动化研究所 | Automatic intersection management and control system |
| CN104754011A (en) * | 2013-12-31 | 2015-07-01 | 中国移动通信集团公司 | Internet of Vehicles multi-party coordination control method and Internet of Vehicles coordination platform |
| CN104590274A (en) * | 2014-11-26 | 2015-05-06 | 浙江吉利汽车研究院有限公司 | Driving behavior self-adaptation system and method |
| CN104882008A (en) * | 2015-06-03 | 2015-09-02 | 东南大学 | Method for vehicle cooperative control at non-signaled intersection in vehicle networking environment |
Non-Patent Citations (1)
| Title |
|---|
| 面向自驾联网车的交叉口协同控制方法研究;卓福庆;《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》;20170215;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109727470A (en) | 2019-05-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109727470B (en) | Complex scene passing decision method for distributed intelligent network-connected automobile intersection | |
| CN107798861B (en) | A kind of vehicle cooperative formation driving method and system | |
| CN104269070B (en) | Active vehicle safety pre-warning method and safety pre-warning system with same applied | |
| CN110674565A (en) | An in-the-loop simulation method and platform for a vehicle-road collaborative system | |
| CN113593273A (en) | No-signal control road intersection collision early warning method based on V2I communication | |
| CN108986471A (en) | Intersection vehicles bootstrap technique under the conditions of mixed traffic | |
| CN105894842A (en) | Intelligent traffic signal lamp system | |
| CN107564310A (en) | A kind of bus or train route interacted system and method based on the processing of Traffic Information cloud | |
| Kim et al. | A modeling framework for connectivity and automation co-simulation | |
| CN111724585B (en) | Signal timing sequence control method and optimization method for signal-controlled intersection | |
| CN115565390B (en) | Intelligent network-connected automobile multi-lane queue traffic control method, system and computer readable storage medium | |
| Pan et al. | Research on the behavior decision of connected and autonomous vehicle at the unsignalized intersection | |
| CN114758519A (en) | Vehicle road cooperative automatic driving system based on 5G and V2X intelligent lamp posts | |
| CN110444015A (en) | Intelligent network based on no signal crossroad subregion joins car speed decision-making technique | |
| CN110503832A (en) | Command processing method, device, equipment and medium based on wisdom traffic system | |
| CN117238152A (en) | Safe and efficient scheduling system and method for vehicle-port cooperative mixed traffic | |
| CN115047840A (en) | Signal machine hardware-in-loop simulation test system and method | |
| CN206194129U (en) | Device based on induced tactful guide car of vehicle velocity does not stop and to pass through crossing | |
| CN115631644A (en) | Vehicle traffic control method, device, electronic equipment and computer storage medium | |
| CN117576904B (en) | A method for guiding other vehicles to efficiently pass through traffic lights by intelligent networked vehicles | |
| CN115250284B (en) | High-speed operation area safety control system based on internet of things cloud service | |
| Alanazi et al. | Improving the performance of unsignalized T-intersections within CAVs mixed traffic | |
| Ding et al. | Research on safety lane change warning method based on potential angle collision point | |
| CN117533314A (en) | Vehicle operation control method and system based on driving environment change | |
| CN117831294A (en) | Bus ecological speed guiding method based on coupling timing information |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20191231 Address after: 100083 401a, floor 4, building 6, yard 1, Zhongguancun East Road, Haidian District, Beijing Applicant after: Beijing Chaoxing Future Technology Co., Ltd Address before: 100084 Haidian District Tsinghua Yuan Beijing No. 1 Applicant before: Tsinghua University |
|
| TA01 | Transfer of patent application right | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |