CN112967516B - Global dynamic path planning method for matching of key parameters of quick parking lot end with whole vehicle - Google Patents

Global dynamic path planning method for matching of key parameters of quick parking lot end with whole vehicle Download PDF

Info

Publication number
CN112967516B
CN112967516B CN202110151518.5A CN202110151518A CN112967516B CN 112967516 B CN112967516 B CN 112967516B CN 202110151518 A CN202110151518 A CN 202110151518A CN 112967516 B CN112967516 B CN 112967516B
Authority
CN
China
Prior art keywords
vehicle
parking
parking space
parking lot
state
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.)
Expired - Fee Related
Application number
CN202110151518.5A
Other languages
Chinese (zh)
Other versions
CN112967516A (en
Inventor
张永华
朱金波
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Wuhu Ruite Microelectronics Co ltd
Original Assignee
Wuhu Bolatu Information Technology Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Wuhu Bolatu Information Technology Co ltd filed Critical Wuhu Bolatu Information Technology Co ltd
Priority to CN202110151518.5A priority Critical patent/CN112967516B/en
Publication of CN112967516A publication Critical patent/CN112967516A/en
Application granted granted Critical
Publication of CN112967516B publication Critical patent/CN112967516B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • GPHYSICS
    • G08SIGNALLING
    • G08GTRAFFIC CONTROL SYSTEMS
    • G08G1/00Traffic control systems for road vehicles
    • G08G1/09Arrangements for giving variable traffic instructions
    • G08G1/0962Arrangements for giving variable traffic instructions having an indicator mounted inside the vehicle, e.g. giving voice messages
    • G08G1/0968Systems involving transmission of navigation instructions to the vehicle
    • G08G1/096805Systems involving transmission of navigation instructions to the vehicle where the transmitted instructions are used to compute a route
    • GPHYSICS
    • G08SIGNALLING
    • G08GTRAFFIC CONTROL SYSTEMS
    • G08G1/00Traffic control systems for road vehicles
    • G08G1/01Detecting movement of traffic to be counted or controlled
    • G08G1/017Detecting movement of traffic to be counted or controlled identifying vehicles
    • G08G1/0175Detecting movement of traffic to be counted or controlled identifying vehicles by photographing vehicles, e.g. when violating traffic rules
    • GPHYSICS
    • G08SIGNALLING
    • G08GTRAFFIC CONTROL SYSTEMS
    • G08G1/00Traffic control systems for road vehicles
    • G08G1/14Traffic control systems for road vehicles indicating individual free spaces in parking areas

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Remote Sensing (AREA)
  • Traffic Control Systems (AREA)

Abstract

本发明属于车辆自动驾驶领域,公开了快速停车场端关键参数与整车匹配全局动态路径规划方法,包括五个步骤:参数和车位状态录入场端云平台;通过车牌识别出车辆型号并给每辆车匹配出目标车位以及相应的全局路径;通过雷达设备感知车辆最新定位信息;根据监测的最新车位状态和车辆最新定位点位置,利用场端云平台规划出最新定位点对应的最优全局路径以及目标车位;根据规划出的路径能够选择最优的停车路线,重复步骤,直至完成泊车任务。本发明通过场端云平台与车载通讯设备的信息交互,实时获取停车场内的车型参数及运动信息,并通过场端云平台给每辆有停车任务的车辆实时规划出一条全局最优的停车轨迹,使得不同型号车辆总能尽快完成停车任务。

Figure 202110151518

The invention belongs to the field of automatic driving of vehicles, and discloses a global dynamic path planning method for matching key parameters of a fast parking lot end with a whole vehicle, comprising five steps: entering parameters and parking space status into a cloud platform of the parking lot; The vehicle matches the target parking space and the corresponding global path; the latest positioning information of the vehicle is sensed through the radar device; according to the latest monitoring parking space status and the position of the latest vehicle positioning point, the field cloud platform is used to plan the optimal global path corresponding to the latest positioning point and the target parking space; the optimal parking route can be selected according to the planned path, and the steps are repeated until the parking task is completed. The present invention obtains the vehicle parameters and motion information in the parking lot in real time through the information interaction between the yard cloud platform and the vehicle-mounted communication device, and plans a globally optimal parking for each vehicle with a parking task in real time through the yard cloud platform. trajectories, so that different types of vehicles can always complete the parking task as soon as possible.

Figure 202110151518

Description

快速停车场端关键参数与整车匹配全局动态路径规划方法Global dynamic path planning method for matching key parameters of fast parking lot with vehicle

技术领域technical field

本发明涉及快速停车场端关键参数与整车匹配全局动态路径规划方法,属于车辆自动驾驶技术领域。The invention relates to a global dynamic path planning method for matching key parameters of a fast parking lot terminal with a whole vehicle, and belongs to the technical field of vehicle automatic driving.

背景技术Background technique

近些年,随着车辆智能化水平的逐步提高,自动驾驶技术在很多场景下得以初步实现,如自适应巡航ACC,自动紧急刹车AEB,自动泊车等。其中ACC和AEB由于在复杂的城市交通道路下,依然有很多工况难以应付,给驾驶员带来的实际体验往往不够理想,甚至在极限工况下会带来一些碰撞危险。而停车场内的自动泊车技术由于是在封闭的场景下,很少有行人及其他交通参与者。此外,场内车辆速度较低,基本保持在10km/h以下,并且有场内云平台给每辆车发布指令,进行协同控制,为停车场内的自动泊车系统带来了相当多的便利,使得停车场内自动泊车技术变得更有前景,并且更加迫切。In recent years, with the gradual improvement of the level of vehicle intelligence, autonomous driving technology has been initially realized in many scenarios, such as adaptive cruise ACC, automatic emergency braking AEB, and automatic parking. Among them, ACC and AEB still have many working conditions that are difficult to cope with under complex urban traffic roads, and the actual experience brought to the driver is often not ideal, and even brings some collision dangers under extreme working conditions. As the automatic parking technology in the parking lot is in a closed scene, there are few pedestrians and other traffic participants. In addition, the speed of vehicles in the field is relatively low, which is basically kept below 10km/h, and there is a cloud platform in the field to issue instructions to each vehicle for coordinated control, which brings considerable convenience to the automatic parking system in the parking lot. , making the automatic parking technology in the parking lot more promising and more urgent.

发明内容SUMMARY OF THE INVENTION

针对现有技术的不足,本发明提供了快速停车场端关键参数与整车匹配全局动态路径规划方法,解决了不同型号车辆无法尽快完成停车任务的问题。In view of the deficiencies of the prior art, the present invention provides a global dynamic path planning method for matching the key parameters of the fast parking lot end with the whole vehicle, and solves the problem that vehicles of different types cannot complete the parking task as soon as possible.

本发明为解决其技术问题采用如下技术方案:The present invention adopts following technical scheme for solving its technical problem:

本发明所述的快速停车场端关键参数与整车匹配全局动态路径规划方法,包括以下步骤:The global dynamic path planning method for matching the key parameters of the fast parking lot terminal with the whole vehicle according to the present invention includes the following steps:

步骤一:将停车场端关键参数及当前车位状态实时录入场端云平台;Step 1: Enter the key parameters of the parking lot and the current parking space status into the cloud platform of the parking lot in real time;

步骤二:当每辆车通过停车场入口时,场端云平台能通过车牌识别出车辆型号,并结合停车场当前车位状态利用强化学习的方法给每辆车匹配出目标车位以及相应的全局路径;Step 2: When each vehicle passes through the entrance of the parking lot, the on-site cloud platform can identify the vehicle model through the license plate, and use the reinforcement learning method to match each vehicle with the target parking space and the corresponding global path in combination with the current parking space status of the parking lot. ;

步骤三:当车辆具体进入停车场内时,通过车载毫米波雷达,摄像头,并结合停车场内地图,感知到车辆的位置及所处的最新定位点信息;Step 3: When the vehicle enters the parking lot, the vehicle's location and the latest positioning point information are sensed through the on-board millimeter-wave radar and camera combined with the map in the parking lot;

步骤四:场端云平台根据停车场内监测的最新车位状态,以及车辆所处的最新定位点位置,重新利用强化学习的方法实时规划出最新定位点对应的最优全局路径以及目标车位;Step 4: According to the latest parking space status monitored in the parking lot and the latest positioning point position of the vehicle, the field cloud platform reuses the reinforcement learning method to plan the optimal global path corresponding to the latest positioning point and the target parking space in real time;

步骤五:根据规划出的路径,将跟踪控制指令发送到底层控制器和停车场云台,从而进入下一时刻,并重复步骤三到步骤五,直至完成泊车任务。Step 5: According to the planned path, send the tracking control command to the bottom controller and the parking lot PTZ, so as to enter the next moment, and repeat steps 3 to 5 until the parking task is completed.

进一步,在所述步骤一中,将停车场端关键参数及当前车位状态实时录入场端云平台时,具体包括以下内容:Further, in the first step, when the key parameters of the parking lot and the current parking space status are entered into the cloud platform of the parking lot in real time, the specific content includes the following:

1)停车场端的关键参数包括车位分区A,B,C,D,E等,其中不同分区可以停靠不同类型的车辆,如A区可以停靠轴距超过2m的大车,B区则停靠略小一些的车辆,具体如下,1) The key parameters of the parking lot include parking space zones A, B, C, D, E, etc., in which different zones can park different types of vehicles. For example, zone A can park large vehicles with a wheelbase of more than 2m, and zone B can park slightly smaller Some of the vehicles are as follows,

A={a1,a2,…,ap}B={b1,b2,…,bq}A={a 1 ,a 2 ,...,a p }B={b 1 ,b 2 ,...,b q }

其中ai,bi为具体车位情况,如果当前车位为空则为1,已经停车则为0;Among them, a i , b i are the specific parking space conditions, if the current parking space is empty, it is 1, and it is 0 if it has been parked;

2)停车场端的关键参数包括场内预先标记的定位点,代表场内道路特征信息,每个定位点作为一个状态s,将所有定位点标记后可得到n个状态:S={s1,s2,…,sn};2) The key parameters of the parking lot include the pre-marked positioning points in the field, which represent the road feature information in the field. Each positioning point is used as a state s. After marking all the positioning points, n states can be obtained: S={s 1 , s 2 ,…,s n };

3)关键定位点下车位信息,要使得车辆停入某个车位,需要经过某个定位点才能停过去,具体表示如下,3) The parking space information of the key positioning point, to make the vehicle park in a certain parking space, it needs to pass a certain positioning point to stop, the specific expression is as follows,

Figure BDA0002932212800000021
Figure BDA0002932212800000021

其中sm为关键定位点,ai,ai+1,bj,bj+1,cj,cj+1,…为经过该定位点可直接到达的车位,如想进入a1-a5这五个车位,则必须经过s3或者s6才能到达,从而表示如Among them, s m is the key positioning point, a i , a i+1 , b j , b j+1 , c j , c j+1 , ... are the parking spaces that can be directly reached through the positioning point, if you want to enter a 1 - The five parking spaces of a 5 must be reached through s 3 or s 6 , which means that if

Figure BDA0002932212800000022
Figure BDA0002932212800000022

4)定义定位点与定位点之间的动作及转移关系,具体车辆到达每个定位点可以进行4个基本动作A1~A4,其中A1代表向前行驶,A2代表向左转弯,A3代表向右转弯,A4代表静止,对应的状态转移关系可由停车场内结构参数确定,如在S6处采取动作A1则到达S5;采取动作A3则到达S4;采取其它动作还到达本身状态S6,将所有状态对应的转移关系都表征出来可以得到一个状态转移矩阵T。4) Define the action and transfer relationship between the positioning point and the positioning point. The specific vehicle can perform 4 basic actions A1 to A4 when it reaches each positioning point, where A1 means driving forward, A2 means turning left, and A3 means right Turning, A4 stands for static, and the corresponding state transition relationship can be determined by the structural parameters in the parking lot. For example, if action A1 is taken at S6, it will reach S5; if action A3 is taken, it will reach S4; if other actions are taken, it will also reach its own state S6, and all states correspond to The transition relationships of , are characterized, and a state transition matrix T can be obtained.

进一步,在所述步骤二中,利用强化学习的方法给每辆车匹配出目标车位以及相应的全局路径,具体步骤如下:Further, in the second step, the method of reinforcement learning is used to match the target parking space and the corresponding global path for each vehicle, and the specific steps are as follows:

1)当识别出车型后,即可根据车型安排出对应的分区,此时该分区内所有的空位都可能成为最终的目标车位,将这些空车位所属的定位点提取出来,这些定位点也就成为全局路径的最终定位点,可表示如下:1) After the vehicle type is identified, the corresponding partition can be arranged according to the vehicle type. At this time, all the vacant spaces in the partition may become the final target parking spaces, and the positioning points to which these empty parking spaces belong are extracted, and these positioning points are also becomes the final anchor point of the global path, which can be expressed as follows:

Figure BDA0002932212800000023
Figure BDA0002932212800000023

其中Sf为最终定位点集合,

Figure BDA0002932212800000024
为全局路径可能经过的最终定位点;where S f is the final set of positioning points,
Figure BDA0002932212800000024
is the final positioning point that the global path may pass through;

2)建立全局路径的强化学习过程,其中状态集:S={s1,s2,…,sn},动作集Action={A1,A2,A3,A4},及状态转移矩阵T都已经在2.2,2.4中建立,对应的回报函数可以表示如下:2) The reinforcement learning process of establishing a global path, in which the state set: S={s 1 , s 2 ,...,s n }, the action set Action={A 1 , A 2 , A 3 , A 4 }, and the state transition The matrix T has been established in 2.2 and 2.4, and the corresponding reward function can be expressed as follows:

Figure BDA0002932212800000025
Figure BDA0002932212800000025

其中st+1为在状态St处采取动作At达到的下一状态,由状态转移矩阵得到:st+1=T(st,At),该回报代表当下一状态到达最终定位点时,回报会特别大;如果没有到达,则付出一定的距离成本;Among them, s t+1 is the next state reached by taking action A t at state S t , which is obtained from the state transition matrix: s t+1 =T(s t , A t ), the return represents when the next state reaches the final positioning When the point is reached, the reward will be particularly large; if it is not reached, a certain distance cost will be paid;

3)建立Q学习迭代方程:3) Establish the Q-learning iterative equation:

Figure BDA0002932212800000034
Figure BDA0002932212800000034

其中α为学习率,γ为遗忘因子;where α is the learning rate and γ is the forgetting factor;

4)随机初始化状态St,通过迭代方程不断学习,当状态st+1∈Sf时即进入下一循环,循环多次收敛以后,即可得到从当前定位点s0到目标车位最终定位点Sf对应的全局路径:4) Randomly initialize the state S t , and continuously learn through the iterative equation. When the state s t+1 ∈ S f , the next cycle is entered. After the cycle converges for many times, the final positioning from the current positioning point s 0 to the target parking space can be obtained. The global path corresponding to point S f :

s0 a0 s1 a1 … sf s 0 a 0 s 1 a 1 … s f

5)到达最终定位点Sf后,选择距离该定位点最近的空车位作为目标车位即可。5) After reaching the final positioning point S f , select the empty parking space closest to the positioning point as the target parking space.

进一步,所述步骤四中重新利用强化学习的方法实时规划出最优全局路径以及目标车位时,具体步骤如下:Further, when reusing the reinforcement learning method in the step 4 to plan the optimal global path and the target parking space in real time, the specific steps are as follows:

1)根据停车场内监测的最新车位状态,需要对3.1)中的最终定位点集合进行更新,得到当前t时刻对应的新的

Figure BDA0002932212800000031
1) According to the latest parking space status monitored in the parking lot, it is necessary to update the final set of positioning points in 3.1) to obtain a new corresponding to the current time t.
Figure BDA0002932212800000031

2)重复3.2)-3.3)的强化学习过程,重新进行迭代训练,当状态

Figure BDA0002932212800000032
时进入下一循环,循环多次收敛后,即可得到车辆所在的最新定位点到更新后的最终定位点的全局路径:
Figure BDA0002932212800000033
2) Repeat the reinforcement learning process of 3.2)-3.3), and re-train iteratively, when the state
Figure BDA0002932212800000032
When entering the next loop, after the loop converges for many times, the global path from the latest positioning point where the vehicle is located to the updated final positioning point can be obtained:
Figure BDA0002932212800000033

与现有技术相比,本发明具有的有益效果是:Compared with the prior art, the present invention has the following beneficial effects:

通过场端云平台与车载通讯设备的信息交互,实时获取停车场内的车型参数及运动信息,并通过场端云平台给每辆有停车任务的车辆实时规划出一条全局最优的停车轨迹,使得不同型号车辆总能高效快速尽快完成停车任务。Through the information interaction between the field cloud platform and the vehicle communication equipment, the vehicle parameters and motion information in the parking lot are obtained in real time, and a globally optimal parking trajectory is planned in real time for each vehicle with a parking task through the field cloud platform. So that different types of vehicles can always complete the parking task efficiently and quickly as soon as possible.

附图说明Description of drawings

图1是本发明停车场关键参数。Fig. 1 is the key parameter of the parking lot of the present invention.

具体实施方式Detailed ways

为了使本发明的目的、技术方案及优点更加清楚明白,以下结合附图及实施案例对本发明进行深入地详细说明。应当理解,此处所描述的具体实施案例仅仅用以解释本发明,并不用于限定发明。In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be described in detail below with reference to the accompanying drawings and implementation cases. It should be understood that the specific embodiments described herein are only used to explain the present invention, but not to limit the invention.

如图1所示,本发明所述的快速停车场端关键参数与整车匹配全局动态路径规划方法,包括以下步骤:As shown in Figure 1, the global dynamic path planning method for matching the key parameters of the fast parking lot terminal with the whole vehicle according to the present invention includes the following steps:

步骤一:将停车场端关键参数及当前车位状态实时录入场端云平台;Step 1: Enter the key parameters of the parking lot and the current parking space status into the cloud platform of the parking lot in real time;

步骤二:当每辆车通过停车场入口时,场端云平台能通过车牌识别出车辆型号,并结合停车场当前车位状态利用强化学习的方法给每辆车匹配出目标车位以及相应的全局路径;Step 2: When each vehicle passes through the entrance of the parking lot, the on-site cloud platform can identify the vehicle model through the license plate, and use the reinforcement learning method to match each vehicle with the target parking space and the corresponding global path in combination with the current parking space status of the parking lot. ;

步骤三:当车辆具体进入停车场内时,通过车载毫米波雷达,摄像头,并结合停车场内地图,感知到车辆的位置及所处的最新定位点信息;Step 3: When the vehicle enters the parking lot, the vehicle's location and the latest positioning point information are sensed through the on-board millimeter-wave radar and camera combined with the map in the parking lot;

步骤四:场端云平台根据停车场内监测的最新车位状态,以及车辆所处的最新定位点位置,重新利用强化学习的方法实时规划出最新定位点对应的最优全局路径以及目标车位;Step 4: According to the latest parking space status monitored in the parking lot and the latest positioning point position of the vehicle, the field cloud platform reuses the reinforcement learning method to plan the optimal global path corresponding to the latest positioning point and the target parking space in real time;

步骤五:根据规划出的路径,将跟踪控制指令发送到底层控制器和停车场云台,从而进入下一时刻,并重复步骤三到步骤五,直至完成泊车任务。Step 5: According to the planned path, send the tracking control command to the bottom controller and the parking lot PTZ, so as to enter the next moment, and repeat steps 3 to 5 until the parking task is completed.

进一步,在所述步骤一中,将停车场端关键参数及当前车位状态实时录入场端云平台时,具体包括以下内容:Further, in the first step, when the key parameters of the parking lot and the current parking space status are entered into the cloud platform of the parking lot in real time, the specific content includes the following:

1)停车场端的关键参数包括车位分区A,B,C,D,E等,其中不同分区可以停靠不同类型的车辆,如A区可以停靠轴距超过2m的大车,B区则停靠略小一些的车辆,具体如下,1) The key parameters of the parking lot include parking space zones A, B, C, D, E, etc., in which different zones can park different types of vehicles. For example, zone A can park large vehicles with a wheelbase of more than 2m, and zone B can park slightly smaller Some of the vehicles are as follows,

A={a1,a2,…,ap}B={b1,b2,…,bq}A={a 1 ,a 2 ,...,a p }B={b 1 ,b 2 ,...,b q }

其中ai,bi为具体车位情况,如果当前车位为空则为1,已经停车则为0;Among them, a i , b i are the specific parking space conditions, if the current parking space is empty, it is 1, and it is 0 if it has been parked;

2)停车场端的关键参数包括场内预先标记的定位点,代表场内道路特征信息,每个定位点作为一个状态s,将所有定位点标记后可得到n个状态:S={s1,s2,…,sn};2) The key parameters of the parking lot include the pre-marked positioning points in the field, which represent the road feature information in the field. Each positioning point is used as a state s. After marking all the positioning points, n states can be obtained: S={s 1 , s 2 ,…,s n };

3)关键定位点下车位信息,要使得车辆停入某个车位,需要经过某个定位点才能停过去,具体表示如下,3) The parking space information of the key positioning point, to make the vehicle park in a certain parking space, it needs to pass a certain positioning point to stop, the specific expression is as follows,

Figure BDA0002932212800000041
Figure BDA0002932212800000041

其中sm为关键定位点,ai,ai+1,bj,bj+1,cj,cj+1,…为经过该定位点可直接到达的车位,如要想进入a1-a5这五个车位,则必须经过s3或者s6才能到达,从而表示如下,Among them, s m is the key positioning point, a i , a i+1 , b j , b j+1 , c j , c j+1 , ... are the parking spaces that can be directly reached through the positioning point, if you want to enter a 1 -a 5 these five parking spaces, you must go through s 3 or s 6 to arrive, which is expressed as follows,

Figure BDA0002932212800000042
Figure BDA0002932212800000042

4)定义定位点与定位点之间的动作及转移关系,具体车辆到达每个定位点可以进行4个基本动作A1~A4,其中A1代表向前行驶,A2代表向左转弯,A3代表向右转弯,A4代表静止,对应的状态转移关系可由停车场内结构参数确定,如在S6处采取动作A1则到达S5;采取动作A3则到达S4;采取其它动作还到达本身状态S6,将所有状态对应的转移关系都表征出来可以得到一个状态转移矩阵T。4) Define the action and transfer relationship between the positioning point and the positioning point. The specific vehicle can perform 4 basic actions A1 to A4 when it reaches each positioning point, where A1 means driving forward, A2 means turning left, and A3 means right Turning, A4 stands for static, and the corresponding state transition relationship can be determined by the structural parameters in the parking lot. For example, if action A1 is taken at S6, it will reach S5; if action A3 is taken, it will reach S4; if other actions are taken, it will also reach its own state S6, and all states correspond to The transition relationships of , are characterized, and a state transition matrix T can be obtained.

进一步,在所述步骤二中,利用强化学习的方法给每辆车匹配出目标车位以及相应的全局路径,具体步骤如下:Further, in the second step, the method of reinforcement learning is used to match the target parking space and the corresponding global path for each vehicle, and the specific steps are as follows:

1)当识别出车型后,即可根据车型安排出对应的分区,此时该分区内所有的空位都可能成为最终的目标车位,将这些空车位所属的定位点提取出来,这些定位点也就成为全局路径的最终定位点,可表示如下:1) After the vehicle type is identified, the corresponding partition can be arranged according to the vehicle type. At this time, all the vacant spaces in the partition may become the final target parking spaces, and the positioning points to which these empty parking spaces belong are extracted, and these positioning points are also becomes the final anchor point of the global path, which can be expressed as follows:

Figure BDA0002932212800000051
Figure BDA0002932212800000051

其中Sf为最终定位点集合,

Figure BDA0002932212800000052
为全局路径可能经过的最终定位点;where S f is the final set of positioning points,
Figure BDA0002932212800000052
is the final positioning point that the global path may pass through;

2)建立全局路径的强化学习过程,其中状态集:S={s1,s2,…,sn},动作集Action={A1,A2,A3,A4},及状态转移矩阵T都已经在2.2,2.4中建立,对应的回报函数可以表示如下:2) The reinforcement learning process of establishing a global path, in which the state set: S={s 1 , s 2 ,...,s n }, the action set Action={A 1 , A 2 , A 3 , A 4 }, and the state transition The matrix T has been established in 2.2 and 2.4, and the corresponding reward function can be expressed as follows:

Figure BDA0002932212800000053
Figure BDA0002932212800000053

其中st+1为在状态St处采取动作At达到的下一状态,由状态转移矩阵得到:st+1=T(st,At),该回报代表当下一状态到达最终定位点时,回报会特别大;如果没有到达,则付出一定的距离成本;Among them, s t+1 is the next state reached by taking action A t at state S t , which is obtained from the state transition matrix: s t+1 =T(s t , A t ), the return represents when the next state reaches the final positioning When the point is reached, the reward will be particularly large; if it is not reached, a certain distance cost will be paid;

3)建立Q学习迭代方程:3) Establish the Q-learning iterative equation:

Figure BDA0002932212800000054
Figure BDA0002932212800000054

其中α为学习率,γ为遗忘因子;where α is the learning rate and γ is the forgetting factor;

4)随机初始化状态St,通过迭代方程不断学习,当状态st+1∈Sf时即进入下一循环,循环多次收敛以后,即可得到从当前定位点s0到目标车位最终定位点Sf对应的全局路径:4) Randomly initialize the state S t , and continuously learn through the iterative equation. When the state s t+1 ∈ S f , the next cycle is entered. After the cycle converges for many times, the final positioning from the current positioning point s 0 to the target parking space can be obtained. The global path corresponding to point S f :

s0 a0 s1 a1 … sf s 0 a 0 s 1 a 1 … s f

5)到达最终定位点Sf后,选择距离该定位点最近的空车位作为目标车位即可。5) After reaching the final positioning point S f , select the empty parking space closest to the positioning point as the target parking space.

进一步,所述步骤四中重新利用强化学习的方法实时规划出最优全局路径以及目标车位时,具体步骤如下:Further, when reusing the reinforcement learning method in the step 4 to plan the optimal global path and the target parking space in real time, the specific steps are as follows:

1)根据停车场内监测的最新车位状态,需要对3.1)中的最终定位点集合进行更新,得到当前t时刻对应的新的

Figure BDA0002932212800000055
1) According to the latest parking space status monitored in the parking lot, it is necessary to update the final set of positioning points in 3.1) to obtain a new corresponding to the current time t.
Figure BDA0002932212800000055

2)重复3.2)-3.3)的强化学习过程,重新进行迭代训练,当状态

Figure BDA0002932212800000056
时进入下一循环,循环多次收敛后,即可得到车辆所在的最新定位点到更新后的最终定位点的全局路径:
Figure BDA0002932212800000057
2) Repeat the reinforcement learning process of 3.2)-3.3), and re-train iteratively, when the state
Figure BDA0002932212800000056
When entering the next loop, after the loop converges for many times, the global path from the latest positioning point where the vehicle is located to the updated final positioning point can be obtained:
Figure BDA0002932212800000057

本专利主要研究自动泊车中的关键技术,主要是考虑停车场端关键参数时各种整车进行动态路径规划的技术,从而实现每辆有停车任务的车辆实时规划出一条全局最优的停车轨迹,使得不同型号车辆总能高效快速尽快完成停车任务。This patent mainly studies the key technologies in automatic parking, mainly the technology of dynamic path planning for various vehicles when the key parameters of the parking lot are considered, so that each vehicle with a parking task can plan a globally optimal parking in real time. trajectories, so that different types of vehicles can always complete the parking task efficiently and quickly as soon as possible.

Claims (2)

1. The global dynamic path planning method for matching the key parameters of the fast parking lot end with the whole vehicle is characterized by comprising the following steps of:
the method comprises the following steps: recording parking lot end key parameters and the current parking stall state into a lot end cloud platform in real time;
step two: when each vehicle passes through the entrance of the parking lot, the field end cloud platform can identify the vehicle model through the license plate, and matches each vehicle with a target parking space and a corresponding global path by using a reinforcement learning method in combination with the current parking space state of the parking lot;
step three: when a vehicle enters a parking lot, the position of the vehicle and the latest positioning point information of the vehicle are sensed through a vehicle-mounted millimeter wave radar and a camera in combination with a map in the parking lot;
step four: the field end cloud platform re-utilizes a reinforcement learning method to plan an optimal global path and a target parking space corresponding to a latest positioning point in real time according to the latest parking space state monitored in the parking field and the latest positioning point position of the vehicle;
step five: according to the planned path, sending a tracking control instruction to the bottom controller and the parking lot holder so as to enter the next moment, and repeating the third step to the fifth step until the parking task is completed;
when the parking lot end key parameters and the current parking space state are recorded into the lot end cloud platform in real time, the method specifically comprises the following contents:
2.1) the key parameters of the parking lot end comprise parking space subareas A, B, C, D and E, wherein different parking spaces stop different types of vehicles, for example, a large vehicle with the wheelbase exceeding 2m stops in the area A, a slightly smaller vehicle stops in the area B, and concretely,
A={a 1 ,a 2 ,…,a p }B={b 1 ,b 2 ,…,b q }
wherein a is i ,b i For the specific parking space condition, if the current parking space is empty, the current parking space is 1, and if the current parking space is already parked, the current parking space is 0;
2.2) key parameters of the parking lot end comprise anchor points marked in advance in the parking lot, representing road characteristic information in the parking lot, taking each anchor point as a state s, and marking all the anchor points to obtain n states:
S={s 1 ,s 2 ,…,s n };
2.3) the information of the vehicle position under the key positioning point, the vehicle can stop after passing through a certain positioning point when the vehicle is stopped in a certain position, which is specifically expressed as follows,
Figure FDA0003684129890000011
wherein s is m Is a key anchor point, a i ,a i+1 ,b j ,b j+1 ,c j ,c j+1 … indicates a parking space directly accessible via the location point, for example, to enter a 1 -a 5 The five parking spaces must pass through s 3 Or s 6 Can be reached, and thus is represented as follows,
Figure FDA0003684129890000012
2.4) defining the action and transfer relation between the positioning points, specifically, the vehicle arrives at each positioning point to perform 4 basic actions A1-A4, wherein A1 represents forward driving, A2 represents leftward turning, A3 represents rightward turning, A4 represents static, the corresponding state transfer relation can be determined by the structural parameters in the parking lot, and if the action A1 is taken at S6, the vehicle arrives at S5; take action A3 to S4; taking other actions and reaching the self state S6, and representing the transfer relations corresponding to all the states to obtain a state transfer matrix T;
a target parking space and a corresponding global path are matched for each vehicle by using a reinforcement learning method, and the method comprises the following specific steps:
3.1) after the vehicle type is identified, arranging a corresponding partition according to the vehicle type, wherein all vacant positions in the partition can become final target parking spaces, extracting positioning points belonging to the vacant parking spaces, and the positioning points become final positioning points of the global path, and can be represented as follows:
Figure FDA0003684129890000021
wherein S f To be the finalA set of anchor points is set up,
Figure FDA0003684129890000022
final anchor points which can be passed by the global path;
3.2) a reinforcement learning process to establish a global path, wherein the state set: s ═ S 1 ,s 2 ,…,s n }, Action set Action ═ a 1 ,A 2 ,A 3 ,A 4 The, and state transition matrix T are already established in 2.2), 2.4), the corresponding return function is expressed as follows:
Figure FDA0003684129890000023
wherein s is t+1 Is in a state S t Take action A t The next state reached, derived from the state transition matrix:
s t+1 =T(s t ,A t ) The return represents that the return is particularly large when the next state reaches the final anchor point; if not, paying a certain distance cost;
3.3) establishing a Q learning iterative equation:
Q(s t ,a t )=Q(s t ,a t )+α[r t +γmax ai Q(s t+1 ,a i )-Q(s t ,a t )]
wherein alpha is the learning rate and gamma is the forgetting factor;
3.4) random initialization State S t Continuously learning through an iterative equation, when the state s t+1 ∈S f Entering the next cycle, and obtaining the current positioning point s after the cycle is converged for multiple times 0 Final positioning point S to target parking space f The corresponding global path:
s 0 →a 0 →s 1 →a 1 →…→s f
3.5) reach the final location S f And then, selecting the empty parking space closest to the positioning point as the target parking space.
2. The method for planning the global dynamic path for matching the key parameters of the fast parking lot end with the whole vehicle according to claim 1, wherein when the optimal global path and the target parking space are planned in real time by the reinforcement learning method in the fourth step, the specific steps are as follows:
4.1) according to the latest parking space state monitored in the parking lot, updating the final positioning point set in 3.1) to obtain a new positioning point corresponding to the current time t
Figure FDA0003684129890000031
4.2) repeating the reinforcement learning process of 3.2) -3.3), and carrying out iterative training again, when the state is the same
Figure FDA0003684129890000032
And then entering the next cycle, and after the cycle is converged for multiple times, obtaining a global path from the latest positioning point where the vehicle is located to the updated final positioning point:
Figure FDA0003684129890000033
CN202110151518.5A 2021-02-03 2021-02-03 Global dynamic path planning method for matching of key parameters of quick parking lot end with whole vehicle Expired - Fee Related CN112967516B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110151518.5A CN112967516B (en) 2021-02-03 2021-02-03 Global dynamic path planning method for matching of key parameters of quick parking lot end with whole vehicle

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110151518.5A CN112967516B (en) 2021-02-03 2021-02-03 Global dynamic path planning method for matching of key parameters of quick parking lot end with whole vehicle

Publications (2)

Publication Number Publication Date
CN112967516A CN112967516A (en) 2021-06-15
CN112967516B true CN112967516B (en) 2022-07-26

Family

ID=76274560

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110151518.5A Expired - Fee Related CN112967516B (en) 2021-02-03 2021-02-03 Global dynamic path planning method for matching of key parameters of quick parking lot end with whole vehicle

Country Status (1)

Country Link
CN (1) CN112967516B (en)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114023102B (en) * 2021-11-16 2024-02-27 厦门市万云智能科技有限公司 Parking space sensing method based on multi-mode detection
CN114913708B (en) * 2022-07-18 2022-10-28 深圳市华睿智兴信息科技有限公司 Parking path guiding system and method for intelligent parking lot
CN117590371B (en) * 2024-01-18 2024-03-29 上海几何伙伴智能驾驶有限公司 Method for realizing global parking space state detection based on 4D millimeter wave imaging radar
CN118781852B (en) * 2024-07-22 2025-04-25 张海林 Parking management method

Family Cites Families (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102522002B (en) * 2011-12-31 2013-11-06 南京九竹科技实业有限公司 All-round intelligent guiding system and method for large parking lot
CN104021375B (en) * 2014-05-29 2017-11-07 银江股份有限公司 A kind of model recognizing method based on machine learning
CN105913685A (en) * 2016-06-25 2016-08-31 上海大学 Video surveillance-based carport recognition and intelligent guide method
US10490066B2 (en) * 2016-12-29 2019-11-26 X Development Llc Dynamic traffic control
JP6936656B2 (en) * 2017-08-10 2021-09-22 三菱地所パークス株式会社 Parking lot monitoring device, parking lot monitoring method, parking lot monitoring program
CN108766017A (en) * 2018-05-08 2018-11-06 上海海事大学 A kind of visualization garage system
CN110466502B (en) * 2018-05-11 2020-12-08 广州汽车集团股份有限公司 Valet parking method and device
CN109147378A (en) * 2018-08-09 2019-01-04 武汉中科通达高新技术股份有限公司 Parking induction method in a kind of parking lot
CN110136481B (en) * 2018-09-20 2021-02-02 初速度(苏州)科技有限公司 Parking strategy based on deep reinforcement learning
CN109803344B (en) * 2018-12-28 2019-10-11 北京邮电大学 A joint construction method of UAV network topology and routing
CN110782696B (en) * 2019-01-25 2022-01-04 长城汽车股份有限公司 Control system for passenger-riding parking
CN109741626B (en) * 2019-02-24 2023-09-29 苏州科技大学 Parking situation prediction method, scheduling method and system for parking lot
CN110288852A (en) * 2019-06-04 2019-09-27 恒大智慧科技有限公司 A kind of parking stall management method and system
CN110738866B (en) * 2019-09-19 2022-02-15 江苏大学 Distributed parking lot parking space planning system
CN112068515A (en) * 2020-08-27 2020-12-11 宁波工程学院 Full-automatic parking lot scheduling method based on deep reinforcement learning

Also Published As

Publication number Publication date
CN112967516A (en) 2021-06-15

Similar Documents

Publication Publication Date Title
CN112967516B (en) Global dynamic path planning method for matching of key parameters of quick parking lot end with whole vehicle
DE102018122459B4 (en) VEHICLE SYSTEM, REMOTE VEHICLE ASSISTANCE SYSTEM AND COMPUTER-IMPLEMENTED METHOD FOR PROVIDING REMOTE ASSISTANCE
US20240320466A1 (en) Systems and Methods for Generating Motion Forecast Data for Actors with Respect to an Autonomous Vehicle and Training a Machine Learned Model for the Same
US11131993B2 (en) Methods and systems for trajectory forecasting with recurrent neural networks using inertial behavioral rollout
US12073725B2 (en) Systems and methods for coordinating an arrival of one or more assets at a location
CN110673602B (en) Reinforced learning model, vehicle automatic driving decision method and vehicle-mounted equipment
CN107886750B (en) Unmanned automobile control method and system based on beyond-visual-range cooperative cognition
DE102018115263A1 (en) SYSTEMS AND METHOD FOR PREDICTING TRAFFIC PATTERNS IN AN AUTONOMOUS VEHICLE
DE102018113007A1 (en) TRAJECTOR PLANNER FOR AUTONOMOUS DRIVING BY BÉZIER CURVES
DE102017126877A1 (en) Automated copilot control for autonomous vehicles
DE102019110184A1 (en) AUTONOMOUS SYSTEMS AND METHODS USING DRIVER LEARNING BASED ON NEURAL NETWORKS ON TOKENIZED SENSOR INPUTS
DE102019100572A1 (en) SYSTEMS AND METHODS FOR ROUTE PLANNING IN AUTONOMOUS VEHICLES
DE102018129295A1 (en) Systems and methods for mapping lane delays in autonomous vehicles
DE102019110759A1 (en) SYSTEM AND METHOD FOR CONTROLLING AN AUTONOMIC VEHICLE
DE102018129048A1 (en) SYSTEMS AND METHOD FOR DETERMINING SAFETY EVENTS FOR AN AUTONOMOUS VEHICLE
DE102017126021A1 (en) PROCESSOR-IMPLEMENTED SYSTEMS AND METHOD FOR AUTOMATED DRIVING
DE102018119469A1 (en) SYSTEM AND METHOD FOR IMPROVED OBSTACLE DETECTION USING A V2X COMMUNICATION SYSTEM
DE102018104717A1 (en) SYSTEMS AND METHOD FOR PREDICTING VEHICLE DIMENSIONS
DE112018007253T5 (en) Vehicle control device, vehicle control method and program
DE102019113389A1 (en) SYSTEMS AND METHODS FOR A PREDICTIVE ENTITY BEHAVIOR
DE102019108644A1 (en) METHOD AND DEVICE FOR AUTOMATIC LEARNING OF RULES FOR AUTONOMOUS DRIVING
DE102018129101A1 (en) ACTIVE TRANSPORT PARTICIPANTS
DE102021109389A1 (en) ESTIMATION OF A VIRTUAL LANE USING A RECURSIVE SELF-ORGANIZING MAP
US11975726B1 (en) Systems and methods for interaction-based trajectory prediction
DE102019107485A1 (en) MOVEMENT OF AN AUTONOMOUS VEHICLE TO STATIONARY VEHICLES

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
GR01 Patent grant
GR01 Patent grant
TR01 Transfer of patent right
TR01 Transfer of patent right

Effective date of registration: 20230320

Address after: 241002 20th Floor, Building 7, Science and Technology Industrial Park, Wuhu High and New Technology Industrial Development Zone, Yijiang District, Wuhu City, Anhui Province

Patentee after: Wuhu Ruite Microelectronics Co.,Ltd.

Address before: 241002 No.100, Huajin South Road, Yijiang District, Wuhu City, Anhui Province

Patentee before: Wuhu bolatu Information Technology Co.,Ltd.

CF01 Termination of patent right due to non-payment of annual fee
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20220726