A method and device for planning a driving trajectory for an autonomous vehicle

Abstract

The embodiment of the specification discloses a method for planning a driving track for an automatic driving vehicle, comprising the following steps: a cloud control platform acquires vehicle state data sent by the automatic driving vehicle and road environment data sent by a roadside sensing device on a road where the automatic driving vehicle travels; target track information corresponding to the automatic driving vehicle is calculated by using a pre-established neural network model according to the vehicle state data and the road environment data; wherein the track represented by the target track information is a track expected to be traveled by the automatic driving vehicle within a preset time range; the target track information is sent to the automatic driving vehicle, so that the automatic driving vehicle travels along the track represented by the target track information; the problem that a single vehicle intelligent automatic driving is not enough for identifying and reacting to static traffic facilities and static traffic participants is solved, the safety of automatic driving is improved, and the adaptability of the automatic driving vehicle to bad weather and road environment conditions is improved.

Description

Technical Field

[0001] This application relates to the field of autonomous driving technology, specifically to a method and apparatus for planning the driving trajectory of autonomous vehicles. Background Technology

[0002] Autonomous driving technology is a crucial factor influencing the future development of the automotive industry. As autonomous driving technology matures and becomes commercialized, cars will no longer be driving tools that can only be directly controlled by humans. The core value of cars will shift from the transmission system that embodies power and handling to the software system that embodies intelligence, thereby freeing the driver's eyes, hands, and feet.

[0003] Today's autonomous driving technology refers to single-vehicle intelligent autonomous driving technology. During operation, the vehicle needs to autonomously recognize traffic signs, read traffic lights, identify objects on the road, and perform real-time trajectory planning and vehicle control. That is, it controls the actuators of the autonomous vehicle to follow the planned trajectory, driving like a human. A single-vehicle intelligent autonomous driving system consists of autonomous driving hardware and software. The hardware includes LiDAR, millimeter-wave radar, cameras, ultrasonic sensors, GPS positioning devices, chips, and computing platforms. The software includes high-precision mapping systems, high-precision positioning systems, perception systems, decision-making and planning systems, vehicle control systems, and vehicle communication systems. Although single-vehicle intelligent autonomous driving technology has made significant progress and achieved some commercialization success, there is still a long way to go before large-scale widespread application.

[0004] Currently, the safety of single-vehicle intelligent autonomous driving still faces significant challenges, with potential failure risks in certain scenarios. Case studies of numerous autonomous driving safety accidents both domestically and internationally have revealed that single-vehicle sensors struggle to accurately identify stationary traffic facilities and pedestrians, leading to numerous accidents. Furthermore, single-vehicle sensors have relatively large blind spots and insufficient perception range.

[0005] Furthermore, the adaptability of single-vehicle intelligent autonomous driving is limited. In adverse weather and road conditions (rain, snow, fog, backlight, tunnel entrances, etc.), it is not conducive to the continuous operation of autonomous driving. Therefore, current single-vehicle intelligent autonomous driving vehicles can only operate in limited environments.

[0006] Furthermore, single-vehicle intelligent autonomous vehicles need to achieve the highest levels of safety and stability, even exceeding the requirements for ordinary cars. Therefore, these vehicles require more sensors, auxiliary positioning equipment, real-time computing platforms, and supporting hardware and software. This significantly increases the overall vehicle cost, making mass production and application difficult.

[0007] Based on this, this application is hereby submitted to address the problem that single-vehicle intelligent autonomous driving is not capable enough in recognizing and reacting to stationary traffic facilities and stationary traffic participants, thereby improving the safety of autonomous driving and enhancing the adaptability of autonomous vehicles to adverse weather and road conditions.

[0008] This will further enable "lightweighting" of intelligent hardware and software configurations for autonomous vehicles in highway scenarios, and will allow for the rapid realization of intelligentization in mass-produced vehicles. Summary of the Invention

[0009] With the in-depth development of autonomous driving technology, connected autonomous driving technology based on single-vehicle intelligent autonomous driving has emerged, such as cloud-controlled autonomous driving technology. It is a new technical route and a new stage of development for autonomous driving. Cloud-controlled autonomous driving technology fully leverages the advantages of road systems and vehicle intelligence and networking to achieve vehicle-to-vehicle, vehicle-to-road, vehicle-to-cloud, and vehicle-to-human collaboration, ultimately achieving fully automated driving.

[0010] Therefore, embodiments of this specification provide a method and apparatus for planning the driving trajectory of autonomous vehicles, in order to solve the problem that existing single-vehicle intelligent autonomous driving is not capable enough in recognizing and reacting to stationary traffic facilities and stationary traffic participants, thereby improving the safety of autonomous driving and enhancing the adaptability of autonomous vehicles to adverse weather and road conditions.

[0011] Furthermore, through the cloud control platform trajectory planning system in highway scenarios, it is possible to achieve "lightweighting" of intelligent hardware and software configurations for connected autonomous vehicles, enabling rapid realization of intelligent mass-produced vehicles.

[0012] It should be noted that the cloud control platform trajectory planning system is the system for planning the driving trajectory of autonomous vehicles in this invention. The cloud control platform trajectory planning system is also called the cloud control platform trajectory algorithm system. Its function is to obtain the target trajectory information corresponding to the autonomous vehicle by analyzing and calculating the vehicle status data of the autonomous vehicle obtained by the cloud control platform and the road environment data sent by the roadside perception device, and then send the target trajectory information to the autonomous vehicle so that the autonomous vehicle can follow the target trajectory.

[0013] The embodiments in this specification adopt the following technical solutions:

[0014] This specification provides a method for planning the driving trajectory of autonomous vehicles, including:

[0015] The cloud control platform acquires vehicle status data sent by autonomous vehicles;

[0016] The system acquires road environment data sent by roadside sensing devices on the road where the autonomous vehicle is traveling; the road environment data includes driving data of other vehicles and pedestrian data within the driving area of ​​the autonomous vehicle.

[0017] Based on the vehicle status data and the road environment data, a pre-established neural network model is used to calculate the target trajectory information corresponding to the autonomous vehicle; wherein, the neural network model is used to generate the target trajectory information of the autonomous vehicle; the trajectory represented by the target trajectory information is the trajectory that the autonomous vehicle is expected to travel within a preset time range;

[0018] The target trajectory information is sent to the autonomous vehicle so that the autonomous vehicle can follow the trajectory indicated by the target trajectory information.

[0019] This specification also provides an apparatus for planning the driving trajectory of an autonomous vehicle, comprising:

[0020] The acquisition module is used by the cloud control platform to acquire vehicle status data sent by the autonomous vehicle;

[0021] The acquisition module is further configured to acquire road environment data sent by roadside perception devices on the road where the autonomous vehicle is traveling; the road environment data includes driving data of other vehicles and pedestrian data within the driving area of ​​the autonomous vehicle.

[0022] The calculation module is used to calculate the target trajectory information corresponding to the autonomous vehicle using a pre-established neural network model based on the vehicle status data and the road environment data; wherein, the neural network model is used to generate the target trajectory information of the autonomous vehicle; the trajectory represented by the target trajectory information is the trajectory that the autonomous vehicle is expected to travel within a preset time range;

[0023] A sending module is used to send the target trajectory information to the autonomous vehicle so that the autonomous vehicle can follow the trajectory represented by the target trajectory information.

[0024] The above-described at least one technical solution adopted in the embodiments of this specification can achieve the following beneficial effects:

[0025] The system acquires vehicle status data sent by the autonomous vehicle through a cloud control platform; it also acquires road environment data sent by roadside sensing devices on the road where the autonomous vehicle is traveling; the road environment data includes driving data of other vehicles and pedestrian data within the autonomous vehicle's driving area; based on the vehicle status data and the road environment data, a pre-established neural network model is used to calculate the target trajectory information corresponding to the autonomous vehicle; wherein, the neural network model is used to generate the target trajectory information of the autonomous vehicle; the trajectory represented by the target trajectory information is the trajectory that the autonomous vehicle is expected to travel within a preset time range; the target trajectory information is sent to the autonomous vehicle so that the autonomous vehicle can follow the trajectory represented by the target trajectory information; this solves the problem of insufficient recognition and reaction capabilities of single-vehicle intelligent autonomous driving for stationary traffic facilities and stationary traffic participants, improves the safety of autonomous driving, and enhances the adaptability of autonomous vehicles to adverse weather and road environment conditions.

[0026] Furthermore, the embodiments of this application can also achieve "lightweighting" of intelligent hardware and software configuration for autonomous vehicles in highway scenarios, and can quickly realize the intelligence of mass-produced vehicles. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a flowchart illustrating a method for planning the driving trajectory of an autonomous vehicle, as provided in the embodiments of this specification.

[0029] Figure 2 This is a schematic diagram illustrating the basic operation mode of the cloud-controlled tracking autonomous driving system provided in the embodiments of this specification.

[0030] Figure 3 This is a schematic diagram of the cloud data processing architecture of the cloud control platform-based autonomous driving system provided in the embodiments of this specification.

[0031] Figure 4 This is a schematic diagram of an autonomous vehicle in the Frenet coordinate system provided in the embodiments of this specification.

[0032] Figure 5 This is a schematic diagram of path generation and static obstacle avoidance for autonomous vehicles provided in the embodiments of this specification.

[0033] Figure 6 This is a schematic diagram of speed planning and dynamic obstacle avoidance for autonomous vehicles provided in the embodiments of this specification.

[0034] Figure 7 This is a flowchart of a device for planning the driving trajectory of an autonomous vehicle, as provided in the embodiments of this specification. Detailed Implementation

[0035] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application.

[0036] It should be noted that the cloud control platform described in this application is a platform used to remotely provide services to autonomous vehicles. This platform can be a cluster of devices composed of multiple computers or servers, and may also be referred to as an intelligent connected cloud control platform, an autonomous driving control platform, or a cloud control algorithm platform. The cloud control platform includes at least one or more of the following modules: simulation testing module, information interconnection module, data fusion module, standardization module, and cloud collaboration module, etc. This application does not specifically limit the name of the platform; it only uses the cloud control platform as an example to describe the solution. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without creative effort should fall within the scope of protection of this application.

[0037] As mentioned in the background section, existing technologies face several challenges. First, the safety of single-vehicle intelligent autonomous driving faces significant challenges, with potential failure risks in certain scenarios. Case studies of numerous domestic and international autonomous driving safety accidents have revealed that single-vehicle sensors struggle to accurately identify stationary traffic facilities and pedestrians, leading to numerous accidents. Furthermore, single-vehicle sensors have large blind spots and insufficient perception range. Second, the adaptability of single-vehicle intelligent autonomous driving is limited. Adverse weather and road conditions, such as rain, snow, fog, backlighting, and tunnel entrances, hinder continuous operation; therefore, current single-vehicle intelligent autonomous vehicles can only operate in limited environments. Additionally, single-vehicle intelligent autonomous vehicles require extreme safety and stability, exceeding the requirements of ordinary cars. Consequently, existing single-vehicle intelligent autonomous vehicles necessitate more sensors, auxiliary positioning equipment, real-time computing platforms, and supporting hardware and software. These hardware and software components significantly increase the overall vehicle cost, slowing down the mass production of autonomous vehicles.

[0038] To address the above technical problems, this invention provides a method for planning the driving trajectory of an autonomous vehicle, comprising: a cloud control platform acquiring vehicle state data sent by the autonomous vehicle and road environment data sent by roadside perception devices on the road where the autonomous vehicle is traveling; calculating target trajectory information corresponding to the autonomous vehicle using a pre-established neural network model based on the vehicle state data and road environment data; and sending the target trajectory information to the autonomous vehicle so that the autonomous vehicle can follow the trajectory represented by the target trajectory information; wherein the pre-established neural network model includes an autonomous vehicle state verification model 1301, a surrounding traffic participant search model 1302, and a Fre model on the cloud control platform (cloud control algorithm platform). The present application's method comprises a Frenet coordinate system construction model (1303), a Frenet coordinate system representation model for surrounding traffic participants (1304), a Frenet coordinate system representation model for the autonomous vehicle's state (1305), a path generation and static obstacle avoidance model (1306), a path selection model (1307), a speed planning and dynamic obstacle avoidance model (1308), a trajectory generation model (1309), a trajectory verification model (1310), an emergency takeover trajectory planning model (1311), and a trajectory coordinate transformation model (1312). This method addresses the problem of insufficient recognition and reaction capabilities for stationary traffic facilities and participants in single-vehicle intelligent autonomous driving, improving autonomous driving safety and enhancing the adaptability of autonomous vehicles to adverse weather and road conditions. Furthermore, based on the method disclosed in this application, it is also possible to achieve "lightweighting" of intelligent hardware and software configurations for autonomous vehicles in highway scenarios and to rapidly realize the intelligentization of mass-produced vehicles.

[0039] The methods described in this manual will be explained in detail below with specific examples.

[0040] Reference Figure 1 , Figure 1 This is a flowchart illustrating a method for planning the driving trajectory of an autonomous vehicle, as provided in the embodiments of this specification.

[0041] S110: The cloud control platform obtains vehicle status data sent by autonomous vehicles.

[0042] In this step, the autonomous vehicle uses its installed state perception sensors to acquire vehicle state data during its operation. This data includes the vehicle's speed, heading angle, and various bus data. The autonomous vehicle then transmits this state data to the cloud control platform, enabling the platform to access the vehicle's status data along its driving route.

[0043] As an example, in a highway scenario, an autonomous vehicle travels on the highway, acquiring vehicle status data during its journey through its onboard state perception sensors. This data is then uploaded to a cloud control platform in real time. The cloud control platform's real-time computing algorithm receives this data and automatically performs calculations and analyses. The autonomous vehicle sends its own vehicle status information to the cloud control platform in real time, including high-precision positioning information such as latitude and longitude provided by the onboard GPS sensors; and driving status data, including vehicle speed, heading angle, and various bus data detected by its onboard sensors. Based on this real-time vehicle status data, the cloud control platform can analyze the autonomous vehicle's current state and anticipated driving conditions.

[0044] It should be noted that this solution does not impose specific limitations on the sensor devices actually installed in autonomous vehicles. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without creative effort should fall within the scope of protection of this application.

[0045] In practical applications, the environmental perception sensors used in autonomous vehicles mainly include ultrasonic radar, millimeter-wave radar, lidar, single / dual / tri-lens cameras, surround-view cameras, and night vision devices, etc.

[0046] Different sensors have their own advantages and limitations in terms of perception range. The current trend is to use sensor information fusion technology to make up for the shortcomings of individual sensors and improve the safety and reliability of the entire autonomous driving system.

[0047] S120: Obtain road environment data sent by roadside perception devices on the road where the autonomous vehicle is traveling; the road environment data includes driving data of other vehicles and pedestrian data within the driving area of ​​the autonomous vehicle.

[0048] In this step, the cloud control platform acquires vehicle status data sent by the autonomous vehicle, as well as road environment data sent by roadside perception devices on the road where the autonomous vehicle is traveling. The road environment data includes driving data of other vehicles and pedestrian data in the driving area of ​​the autonomous vehicle, and the vehicle types of other vehicles include autonomous vehicles and non-autonomous vehicles. The road environment data may also include road sign information, road obstacle information, etc.

[0049] As an example, the road environment data in this invention is road environment data collected in real time by roadside sensing devices installed on the road and sent to the cloud control platform; this road environment data is associated with the driving area of ​​the autonomous vehicle, and the data upload time is also consistent; for example, when the autonomous vehicle drives automatically in a designated area on the road, it sends its own status data to the cloud control platform, and the roadside sensing devices on the road sense the driving status data of the autonomous vehicle and other vehicles in the environment in real time and uploads it to the cloud control platform. The real-time transmission of the above two types of data is synchronously sent to the same data receiver by different data senders.

[0050] Similar to or in step S110, this section describes the architecture and implementation of a cloud-controlled, trajectory-following autonomous driving system in a highway scenario. In this system, a cloud server sends instructions to the autonomous vehicle, and the vehicle executes these instructions. To facilitate the rapid deployment of connected autonomous driving, a cloud-network integrated platform is used to provide trajectory planning services for autonomous vehicles, based on the existing conditions of these vehicles.

[0051] It should be noted that from the perspective of operators: cloudification of the network, network => cloud; from the perspective of cloud service providers: networking of the cloud, cloud => network; cloud-network integration refers to connecting applications, cloud computing, pipelines, and customers to provide an end-to-end, complete, flexible, and scalable cloud-network integrated solution; in this solution, to achieve the rapid deployment of connected autonomous driving, the cloud-network integrated solution is used as the technical foundation to provide trajectory planning services for autonomous vehicles.

[0052] In a cloud-controlled, line-following autonomous driving system on a highway, the status of surrounding vehicles is perceived using roadside sensing devices. Combined with the environmental and vehicle status data, the autonomous vehicle's trajectory is planned in real time to facilitate its line-following operation. The planned trajectory for the autonomous vehicle over the next few seconds includes the target location, target speed, and target heading angle.

[0053] Furthermore, such as Figure 2 As shown, Figure 2 This is a schematic diagram illustrating the basic operation mode of the cloud-controlled tracking autonomous driving system provided in the embodiments of this specification.

[0054] In this step, the autonomous vehicle reports its own vehicle status data to the cloud control algorithm platform of the real-time computing platform. At the same time, the roadside perception device reports the data of other environmental vehicles and pedestrians within a preset range on the road in which the autonomous vehicle is traveling to the cloud control algorithm platform of the real-time computing platform. Based on the neural network model pre-established in the cloud control platform, the neural network model calculates the target trajectory according to the traffic participant information and the vehicle status data. The target trajectory is then sent to the autonomous vehicle, allowing the autonomous vehicle to follow the target trajectory planned in the cloud.

[0055] It should be noted that traffic participant information includes both stationary and non-stationary traffic participants on the road where the autonomous vehicle is traveling. Stationary traffic participants include stationary environmental vehicles, road signs, directional signs, temporary medians, and temporary roadblocks set up for road construction, etc., on the road where the autonomous vehicle is traveling; stationary vehicles include vehicles parked on the roadside or malfunctioning vehicles. This invention does not specifically limit whether environmental vehicles are malfunctioning or not, but only limits their externally observed stationary state. Non-stationary traffic participants include moving environmental vehicles, pedestrians walking on the road, and animals, etc., on the road where the autonomous vehicle is traveling.

[0056] In recent years, the continuous development of communication technology has made communication between vehicles and infrastructure faster and more reliable. With these advancements, the data collected by roadside perception devices is more accurate and precise, providing more accurate and reliable road environment data for the trajectory planning algorithms of autonomous vehicles on the cloud control algorithm platform, thereby improving the accuracy of autonomous vehicles following the path.

[0057] S130: Based on the vehicle status data and the road environment data, a pre-established neural network model is used to calculate the target trajectory information corresponding to the autonomous vehicle; wherein, the neural network model is used to generate the target trajectory information of the autonomous vehicle; the trajectory represented by the target trajectory information is the trajectory that the autonomous vehicle is expected to travel within a preset time range.

[0058] In this step, based on the vehicle status data and the road environment data, a pre-established neural network model is used to calculate the target trajectory information corresponding to the autonomous vehicle. The pre-established neural network model may include a cloud control algorithm or a cloud-based trajectory planning algorithm from a real-time computing platform on the cloud control platform. Based on the cloud control algorithm, the target trajectory information corresponding to the autonomous vehicle can be obtained. This target trajectory information represents the trajectory that the autonomous vehicle is expected to travel within a preset time range. Furthermore, while the autonomous vehicle receives the planned target trajectory information and travels along the planned trajectory, it continuously sends its own vehicle status data to the cloud control platform. The cloud control algorithm platform can selectively store data from different scenarios, such as overtaking scenarios, lane exit scenarios, and lane merging scenarios.

[0059] Furthermore, based on different types of scene data stored on the cloud control platform, the cloud control platform can train a preset neural network model so that the target trajectory information of the autonomous vehicle calculated by the pre-established neural network model is more consistent with the actual road traffic driving scenario, which is conducive to improving the driving safety of autonomous vehicles and adapting to more driving environments.

[0060] Furthermore, such as Figure 3 As shown, Figure 3 This is a schematic diagram of the cloud data processing architecture of the cloud control platform-based autonomous driving system provided in the embodiments of this specification.

[0061] In the schematic diagram, the cloud data processing architecture is a pre-established neural network model, which includes an autonomous vehicle state verification model 1301, a surrounding traffic participant search model 1302, a Frenet coordinate system construction model 1303, a surrounding traffic participant Frenet coordinate system expression model 1304, an autonomous vehicle state Frenet coordinate system expression model 1305, a path generation and static obstacle avoidance model 1306, a path selection model 1307, a speed planning and dynamic obstacle avoidance model 1308, a trajectory generation model 1309, a trajectory verification model 1310, an emergency takeover trajectory planning model 1311, and a trajectory coordinate transformation model 1312, etc.

[0062] With the deepening development of autonomous driving, cloud-controlled autonomous driving technology based on single-vehicle intelligent autonomous driving has emerged. It represents a new technical route and a new stage of development for autonomous driving. It aims to fully leverage the advantages of intelligent and networked road systems and vehicles to achieve vehicle-to-vehicle, vehicle-to-infrastructure, vehicle-to-cloud, and vehicle-to-human collaboration, thereby realizing fully autonomous driving.

[0063] In a cloud-controlled, line-following autonomous driving system on a highway, the state of vehicles surrounding the autonomous vehicle is perceived using roadside sensing devices. Combined with the environmental and vehicle states, the autonomous vehicle's trajectory is planned in real time to facilitate its line-following operation. A pre-defined neural network computing model plans the autonomous vehicle's trajectory for the next few seconds, including the target position, target speed, and target heading angle.

[0064] As an example, based on the vehicle state data and the road environment data, a pre-established neural network model is used to calculate the target trajectory information corresponding to the autonomous vehicle, such as... Figure 3 The cloud-based data processing architecture of the cloud-controlled tracking autonomous driving system shown allows each model to run sequentially from left to right to calculate the target trajectory information corresponding to the autonomous vehicle. It should be noted that the method of sequentially calculating the various models from left to right in this embodiment is for the convenience of introducing the invention's contents item by item; this calculation method does not constitute a specific limitation on how the pre-established neural network model is used to calculate the target trajectory information corresponding to the autonomous vehicle in this invention.

[0065] In practical applications, the cloud control platform can calculate the target trajectory information corresponding to the autonomous vehicle based on the actual road environment and state data of the autonomous vehicle on its driving road, and call some calculation models (1301-1312) in the neural network model as needed. For example, after the cloud control platform accumulates enough scene data, it can train the neural network model to obtain the trained neural network model. After the autonomous vehicle A safely passes through scene a, which includes road vehicles, the autonomous vehicle B, which is within a preset distance (e.g., 500m) behind vehicle A, also needs to pass through scene a. At this time, the cloud control platform can call the corresponding calculation model according to the actual intersection to obtain the target trajectory information corresponding to vehicle B, without calling the static obstacle avoidance model 1306 to perform the corresponding data calculation.

[0066] Furthermore, the autonomous vehicle state verification model 1301 is invoked. This model is used to determine whether the vehicle's state is within a reasonable range based on vehicle state data and road environment data obtained from the cloud control platform. A reasonable state includes reasonable vehicle state information determined after verifying the position, speed, and heading angle of the autonomous vehicle. For example, it checks whether the state data reported by the autonomous vehicle is within a reasonable range, performs reasonable verification on the vehicle's position, speed, and heading angle, and terminates trajectory planning for the autonomous vehicle if the vehicle has left the highway. Alternatively, it may be necessary to verify whether the autonomous vehicle supports trajectory planning on the cloud control platform based on the reported vehicle state information. If it does not support it, trajectory planning for the vehicle is not performed; if it does support it, a preset calculation model is invoked to further analyze the acquired data related to the vehicle's movement to obtain trajectory planning for that vehicle.

[0067] Furthermore, the surrounding traffic participant search model 1302 is invoked. The surrounding traffic participant search model 1302 is used to determine stationary and / or moving traffic participants within a preset range around the autonomous vehicle based on vehicle status data and road environment data obtained from the cloud control platform.

[0068] As an example, on a highway, stationary and moving traffic participants within a predetermined range of 150 meters ahead and 50 meters behind the autonomous vehicle's direction of travel are identified. The traffic participant information input into the cloud-based trajectory planning algorithm comes from roadside sensing devices, thus saving the cost of developing and implementing sensing functions on the autonomous vehicle. These surrounding traffic participants serve as constraints for trajectory planning; for example, the location of a stationary, disabled vehicle is considered an impassable area in the trajectory planning process.

[0069] Furthermore, a Frenet coordinate system construction model 1303 is used to determine the motion state data of the autonomous vehicle based on the vehicle state data and road environment data obtained from the cloud control platform. The motion state data represents the longitudinal motion state of the autonomous vehicle along the road direction and the lateral motion state perpendicular to the road direction.

[0070] As an example, a model is built in the Frenet coordinate system. Based on the vehicle status data and road environment data obtained from the cloud control platform, the position and shape of the centerline of the lane where the autonomous vehicle is located are extracted from the map. The Frenet coordinate system is then built based on the centerline of the lane where the autonomous vehicle is located for subsequent planning. Through the Frenet coordinate system, the motion of the autonomous vehicle is expressed as longitudinal motion along the road direction and lateral motion perpendicular to the road direction, such as the longitudinal motion along the road direction and lateral motion perpendicular to the road direction that exist during lane changing.

[0071] In practical applications, such as Figure 4 As shown, Figure 4 This is a schematic diagram of an autonomous vehicle in the Frenet coordinate system provided in the embodiments of this specification. Figure 4 This method reflects the motion state of autonomous vehicles, which is expressed as longitudinal motion along the road direction and lateral motion perpendicular to the road direction. Based on this method, the trajectory planning of autonomous vehicles can be decoupled from map information, which includes high-precision map information stored on the cloud control platform.

[0072] Furthermore, the Frenet coordinate system representation model 1304 for surrounding traffic participants is invoked. The Frenet coordinate system representation model 1304 for surrounding traffic participants is used to determine the position information and speed information of traffic participants around the autonomous vehicle based on the vehicle state data and the road environment data. The traffic participants include traffic participants whose own state is moving and / or traffic participants whose own state is stationary.

[0073] As an example, based on the Frenet coordinate system representation model of surrounding traffic participants, the positions and speeds of traffic participants around the autonomous vehicle are expressed in the Frenet coordinate system. These participants are categorized into moving and stationary types, such as moving vehicles and stationary disabled vehicles. The s-coordinate of their projection onto the Frenet coordinate axes is found, and the l-coordinate, representing the lateral offset from the axes, is calculated. The coordinate transformation from latitude and longitude to Frenet coordinates is provided by a map service. Similarly, in the autonomous vehicle state Frenet coordinate system representation module, the position and speed information of the autonomous vehicle are also expressed in the corresponding Frenet coordinate system.

[0074] Furthermore, the Frenet coordinate system representation model 1305 for autonomous vehicle state is invoked. The Frenet coordinate system representation model 1305 for autonomous vehicle state is used to determine the position and speed information of the autonomous vehicle based on the vehicle state data and road environment data obtained from the cloud control platform.

[0075] Furthermore, the path generation and static obstacle avoidance model 1306 is invoked. This model consists of a path generation model and a static obstacle avoidance model. These two models can process the data separately to obtain the corresponding path information and static obstacle avoidance information, or they can be combined and processed simultaneously to obtain both information. This solution does not specify a particular data processing method for the path generation and static obstacle avoidance model 1306.

[0076] As an example, the static obstacle avoidance model is used to determine information about static obstacles around the autonomous vehicle based on the vehicle state data and the road environment data; and to determine the area where the static obstacles affect the autonomous vehicle based on the information about the static obstacles.

[0077] The area affected by static obstacles to autonomous vehicles is represented as an infeasible region in space. In the Frenet coordinate system, the infeasible region is represented as a polygon, and its position and shape can also be obtained. Then, multiple alternative driving path curves are planned to avoid static obstacles. The alternative path curves are piecewise fifth-order polynomial curves, and the l-coordinate is expressed as a fifth-order polynomial of the s-coordinate. There are a total of 6 coefficients in the curve that need to be calculated, as shown in the following mathematical formula 1.

[0078] Mathematical formula 1:

[0079] l(s) = a0 + a1s + a2s 2 +a3s 3 +a4s 4 +a5s 5

[0080] In practical applications, to avoid excessive computation, the number of alternative path curves should not be too large. Therefore, the space is randomly discretized, and points (intersections) are segmented by a fifth-order polynomial curve. Multiple alternative curves (path settings) are generated based on these points, such as... Figure 5 As shown, Figure 5 This is a schematic diagram of path generation and static obstacle avoidance for autonomous vehicles provided in the embodiments of this specification.

[0081] Among the multiple alternative curves, some overlap with polygons representing static obstacles. These are eliminated, leaving multiple paths that autonomous vehicles can travel on.

[0082] Furthermore, the path selection model 1307 is invoked. The path selection model 1307 is used to select a path curve from the drivable path curves as the first path curve, and to determine the first path curve as the target path. The target path represents the path that the autonomous vehicle is expected to travel within a preset time range.

[0083] As an example, the path selection module 1307 selects one path from multiple paths that the autonomous vehicle can travel on, and uses this path as the target path for the autonomous vehicle. The selection of the target path is based on a balance between minimizing the path length and minimizing the path curvature. Furthermore, the path selection module 1307 determines the coefficients of the piecewise fifth-degree polynomial.

[0084] Once the target path is determined, the target trajectory of the autonomous vehicle can be determined based on the speed-time curve of the autonomous vehicle traveling on the target path.

[0085] Furthermore, the speed planning and dynamic obstacle avoidance model 1308 is invoked. This model consists of a speed planning model and a dynamic obstacle avoidance model. The speed planning and dynamic obstacle avoidance models can process the data separately to obtain corresponding speed planning information and dynamic obstacle avoidance information, or they can be combined and processed simultaneously to obtain the corresponding speed planning information and dynamic obstacle avoidance information. This solution does not specify a particular data processing method for the speed planning and dynamic obstacle avoidance model 1308.

[0086] As an example, the dynamic obstacle avoidance model is used to determine information about dynamic obstacles around the autonomous vehicle based on the vehicle state data and the road environment data; and to determine the area where the dynamic obstacles affect the autonomous vehicle based on the information about the dynamic obstacles.

[0087] like Figure 6 As shown, Figure 6This is a schematic diagram of speed planning and dynamic obstacle avoidance for autonomous vehicles provided in the embodiments of this specification. Dynamic obstacles may appear on the path of the autonomous vehicle. The impact of dynamic obstacles on the autonomous vehicle is expressed as an infeasible region in time-location space. The driving routes of the autonomous vehicle d and the surrounding vehicles a, b, and c are shown on the left, and their curves in time-location space are shown on the right. The area where dynamic obstacles affect the autonomous vehicle is the infeasible region. In this case, the size of the dynamic obstacle must be considered, and a certain safety margin must be reserved. Based on the vehicle status information uploaded by the autonomous vehicle and the information of surrounding traffic participants, the speed planning and dynamic obstacle avoidance model 1308 calculates the target speed on the target path corresponding to the autonomous vehicle.

[0088] In practical applications, the optimal driving trajectory can be determined through a speed planning module. The method for calculating the optimal driving trajectory can be: the objective function is to minimize the sum of accelerations j, while the boundary condition is a displacement s constraint; for example... Figure 6 As mentioned above, the displacement s is restricted to avoid dynamic obstacles - environmental vehicles a, b, and c; speeding and stopping are not allowed on highways, so the speed v of vehicles on the road is limited. As shown in the formula in Mathematical Equation 2 below, path planning and speed planning can be decoupled, thereby reducing the amount of computation and enhancing real-time computing capabilities.

[0089] Mathematical formula 2:

[0090]

[0091] st

[0092]

[0093] Furthermore, the trajectory generation model 1309 is invoked. The trajectory generation model 1309 is used to determine the target trajectory information based on the vehicle status data and the road environment data. The target trajectory information is determined by the trajectory generation model based on speed planning, path selection, and a preset time. The target trajectory information includes the vehicle's target position, target heading, target speed, and angular velocity at various times within the preset time.

[0094] As an example, in the trajectory generation model, the results of velocity planning and path selection are combined into the target trajectory. The target trajectory output by the trajectory generation module adds time information to the expected driving path, namely the target position, target heading, target speed, and angular velocity of the autonomous vehicle at various times in the next few seconds.

[0095] Further, trajectory verification model 1310 is invoked. Based on the vehicle dynamics limits of the autonomous vehicle and the position and speed of surrounding vehicles, trajectory verification model 1310 verifies whether the target trajectory information meets preset conditions, obtaining a first verification result. The preset conditions specifically include:

[0096] When the autonomous vehicle travels along the trajectory planned according to the target trajectory information, it can avoid environmental vehicles and avoid conflict with them.

[0097] Furthermore, the control system and actuators of the autonomous vehicle are capable of tracking and executing based on the trajectory represented by the target trajectory information;

[0098] Furthermore, when the autonomous vehicle travels along the trajectory planned by the target trajectory information, the maximum lateral and longitudinal acceleration generated are within the dynamic limits of the vehicle itself.

[0099] In practical applications, in trajectory verification model 1310, the trajectory output by the trajectory generation module is subject to safety verification. The trajectory verification is based on the vehicle's dynamic limits and the position and speed of the surrounding vehicles. That is, the autonomous vehicle needs to avoid conflicts with the surrounding vehicles while driving along the target trajectory. At the same time, the target trajectory issued must be trackable by the autonomous vehicle's control system and actuators, and cannot be unexecutable. The standard for passing the verification is that the maximum lateral and longitudinal accelerations are within the dynamic limits of the autonomous vehicle. If it fails, a new feasible path curve is selected, and subsequent speed planning, trajectory generation, and trajectory verification are performed. If no feasible path curve can pass the trajectory verification, the emergency takeover trajectory planning model is executed.

[0100] Furthermore, the emergency takeover trajectory planning model 1311 is invoked. The emergency takeover trajectory planning model 1311 is used to determine the emergency braking trajectory information of the autonomous vehicle. The trajectory represented by the emergency braking trajectory is the trajectory that the autonomous vehicle is expected to travel within a preset time range when it cannot avoid colliding with other traffic participants.

[0101] In practical applications, the emergency takeover trajectory planning model is used when autonomous vehicles are in dangerous situations where a collision with other road users is possible and unavoidable. By executing the model, an emergency braking trajectory is planned, and the autonomous vehicle receives a takeover alert. Before the driver takes over, the vehicle will complete the braking action according to the emergency braking trajectory. Additionally, if no drivable path curve can be found that passes trajectory verification, the emergency takeover trajectory planning model will also be executed.

[0102] Furthermore, the trajectory coordinate transformation model 1312 is invoked, which is used to convert the trajectory information in the Frenet coordinate system into latitude and longitude coordinates, and to determine the latitude and longitude coordinate system as the latitude and longitude coordinates in the target trajectory information; the latitude and longitude coordinates include longitude, latitude and heading angle.

[0103] In practical applications, a trajectory coordinate transformation model is used to convert trajectories in the Frenet coordinate system into latitude and longitude coordinates for use by autonomous vehicles. These coordinates include longitude, latitude, and heading angle, with true north as zero degrees and clockwise as the positive direction. In other words, the trajectory coordinate transformation model outputs absolute coordinates, allowing autonomous vehicles to control their motors, steering wheels, and other actuators to follow the trajectory issued by the cloud control platform, based on the absolute coordinates such as latitude and longitude and heading angle provided by their installed positioning equipment.

[0104] S140: Send the target trajectory information to the autonomous vehicle so that the autonomous vehicle can follow the trajectory indicated by the target trajectory information.

[0105] In this step, the autonomous vehicle acquires target trajectory information sent by the cloud control platform; based on the autonomous vehicle's own model for vehicle feedback control algorithm, and according to the deviation between the target trajectory information and the vehicle state information, control parameters for the vehicle actuators of the autonomous vehicle are determined; the vehicle actuators include a motor and a steering wheel; the control parameters include steering wheel angle, throttle opening or brake opening; and the vehicle actuators of the autonomous vehicle are controlled to follow the trajectory represented by the target trajectory information according to the control parameters.

[0106] As an example, after receiving the trajectory, the autonomous vehicle can be controlled by calculating the deviation between the target state and the current state and inputting it into a feedback control algorithm, such as a PID control algorithm. Based on the PID control algorithm, control parameters for the autonomous vehicle are obtained, including steering wheel angle, throttle opening or brake opening, etc. The vehicle actuators of the autonomous vehicle control the autonomous vehicle to follow the trajectory represented by the planned target trajectory information according to the corresponding control parameters.

[0107] Based on the above method, this invention realizes the following: planning the driving trajectory of autonomous vehicles in highway scenarios, enhancing the driving safety of autonomous vehicles based on autonomous vehicle status data and real-time road condition data; planning in the cloud and sending the planning results to the autonomous vehicle in the form of a standardized interface, planning the trajectory of the autonomous vehicle in real time, and giving the expected position and speed of the vehicle.

[0108] Furthermore, in the highway scenario trajectory planning system for autonomous vehicles disclosed in this invention, the information on traffic participants surrounding the autonomous vehicle comes from perception data from roadside sensors with good visibility, wide field of view, and broad perception range. Through this source of perception data and the static obstacle avoidance module in the cloud algorithm, the problem of insufficient recognition and reaction capabilities of single-vehicle intelligent autonomous driving to stationary traffic facilities and participants is solved, thus improving the safety of autonomous driving. For the same reason, the cloud-based trajectory planning system for autonomous vehicles provided by this invention has a wider range of normal operation compared to single-vehicle intelligent autonomous driving systems, increasing the adaptability of autonomous vehicles to adverse weather and road conditions.

[0109] Furthermore, the driving trajectory planning system for autonomous vehicles disclosed in this invention deploys radar / visual sensors and other sensing devices used to collect road environment data on road infrastructure, while the planning software and high-precision map data are deployed in the cloud. The autonomous vehicle only needs to follow the trajectory issued by the cloud control platform, eliminating the need for complex autonomous driving hardware and software systems, significantly reducing costs and facilitating mass production applications. In other words, a low-level autonomous vehicle with positioning and tracking capabilities, as well as vehicle control capabilities, capable of following a trajectory issued by the cloud, can become a higher-level autonomous vehicle capable of autonomous driving on highways after connecting to the cloud platform and trajectory planning services. This means that through a cloud-controlled, trajectory-following autonomous driving system in highway scenarios, the intelligent hardware and software configuration of autonomous vehicles can be "lightweighted," enabling rapid intelligentization of mass-produced vehicles.

[0110] Based on the same inventive concept, this specification also provides a device for emergency braking of an autonomous vehicle. (See reference...) Figure 7 , Figure 7 This is a flowchart of a device for planning the driving trajectory of an autonomous vehicle, as provided in the embodiments of this specification.

[0111] like Figure 7 As shown, the apparatus includes:

[0112] Acquisition module 710, the acquisition module 710 is used by the cloud control platform to acquire vehicle status data sent by the autonomous vehicle;

[0113] The acquisition module 710 is further configured to acquire road environment data sent by roadside perception devices on the road where the autonomous vehicle is traveling; the road environment data includes driving data of other vehicles and pedestrian data within the driving area of ​​the autonomous vehicle.

[0114] The calculation module 720 is used to calculate target trajectory information corresponding to the autonomous vehicle using a pre-established neural network model based on the vehicle state data and the road environment data; wherein, the neural network model is used to generate the target trajectory information of the autonomous vehicle; the trajectory represented by the target trajectory information is the trajectory that the autonomous vehicle is expected to travel within a preset time range;

[0115] The sending module 730 is used to send the target trajectory information to the autonomous vehicle so that the autonomous vehicle can follow the trajectory represented by the target trajectory information.

[0116] For a detailed description of the device, please refer to the description of the method above; it will not be repeated here.

[0117] It should also be noted that the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0118] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on its differences from other embodiments. In particular, for... Figure 7 As the apparatus embodiment shown is basically similar to the method embodiment, the description is relatively simple, and relevant parts can be referred to in the description of the method embodiment.

[0119] The above description is merely an embodiment of this specification and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of this application should be included within the scope of the claims of this application.

Claims

1. A method for planning the driving trajectory of an autonomous vehicle, characterized in that, include: The cloud control platform acquires vehicle status data sent by autonomous vehicles; The system acquires road environment data sent by roadside sensing devices on the road where the autonomous vehicle is traveling; the road environment data includes driving data of other vehicles and pedestrian data within the driving area of ​​the autonomous vehicle. Based on the vehicle status data and the road environment data, a pre-established neural network model is used to calculate the target trajectory information corresponding to the autonomous vehicle; the trajectory represented by the target trajectory information is the trajectory that the autonomous vehicle is expected to travel within a preset time range; the neural network model includes an autonomous vehicle status verification model, a surrounding traffic participant search model, a Frenet coordinate system construction model, a surrounding traffic participant Frenet coordinate system expression model, an autonomous vehicle status Frenet coordinate system expression model, a path generation and static obstacle avoidance model, a path selection model, a speed planning and dynamic obstacle avoidance model, a trajectory generation model, a trajectory verification model, an emergency takeover trajectory planning model, and a trajectory coordinate transformation model on the cloud control platform; the cloud control platform calls some calculation models in the neural network model according to needs to calculate the target trajectory information; specifically: The path generation and static obstacle avoidance model consists of a path generation model and a static obstacle avoidance model; based on the static obstacle avoidance model, information about static obstacles around the autonomous vehicle is determined; based on the information about static obstacles, the area where the static obstacles affect the autonomous vehicle is determined; Based on the path selection model, a path curve is selected from the possible drivable path curves as the first path curve, and the first path curve is determined as the target path; the target path is the expected drivability path of the autonomous vehicle within a preset time range. The speed planning and dynamic obstacle avoidance model consists of a speed planning model and a dynamic obstacle avoidance model. Based on the dynamic obstacle avoidance model, information about dynamic obstacles around the autonomous vehicle is determined; and based on the information about dynamic obstacles, the area where the dynamic obstacles affect the autonomous vehicle is determined. Based on the trajectory generation model, the target trajectory information is determined; The target trajectory information is determined based on speed planning, path selection, and a preset time; the target trajectory information includes the vehicle's target position, target heading, target speed, and angular velocity at various moments within the preset time. The results of speed planning and path selection are combined into the target trajectory; The target trajectory information is sent to the autonomous vehicle so that the autonomous vehicle can follow the trajectory indicated by the target trajectory information.

2. The method as described in claim 1, characterized in that, The calculation of the target trajectory information corresponding to the autonomous vehicle using a pre-established neural network model specifically includes: The autonomous vehicle state verification model is used to determine whether the autonomous vehicle is in a reasonable state. The reasonable state includes reasonable vehicle state information determined after verifying the reasonableness of the autonomous vehicle's position, speed, and heading angle.

3. The method as described in claim 1, characterized in that, The process of obtaining the target trajectory information corresponding to the autonomous vehicle using a pre-established neural network computing model specifically includes: Based on the surrounding traffic participant search model, stationary and / or moving traffic participants within a preset range around the autonomous vehicle are identified.

4. The method as described in claim 1, characterized in that, The calculation of the target trajectory information corresponding to the autonomous vehicle using a pre-established neural network model specifically includes: A model is constructed based on the Frenet coordinate system to determine the motion state data of the autonomous vehicle. The motion state data represents the longitudinal motion state of the autonomous vehicle along the road direction and the lateral motion state perpendicular to the road direction. Based on the Frenet coordinate system representation model of the surrounding traffic participants, the position and speed information of the traffic participants around the autonomous vehicle are determined; the traffic participants include traffic participants whose own state is moving, and / or traffic participants whose own state is stationary. Based on the Frenet coordinate system representation model of the autonomous vehicle state, the position and speed information of the autonomous vehicle are determined.

5. The method as described in claim 1, characterized in that, The process of obtaining the target trajectory information corresponding to the autonomous vehicle using a pre-established neural network computing model specifically includes: Based on the trajectory verification model, and according to the vehicle dynamics limits of the autonomous vehicle and the position and speed of surrounding vehicles, the target trajectory information is verified to determine whether it meets preset conditions, thus obtaining a first verification result; the preset conditions specifically include: When the autonomous vehicle travels along the trajectory planned according to the target trajectory information, it can avoid environmental vehicles and avoid conflict with them. Furthermore, the control system and actuators of the autonomous vehicle are capable of tracking and executing based on the trajectory represented by the target trajectory information; Furthermore, when the autonomous vehicle travels along the trajectory planned by the target trajectory information, the maximum lateral and longitudinal acceleration generated are within the dynamic limits of the vehicle itself.

6. The method as described in claim 1, characterized in that, The process of obtaining the target trajectory information corresponding to the autonomous vehicle using a pre-established neural network computing model specifically includes: Based on the emergency takeover trajectory planning model, the emergency braking trajectory information of the autonomous vehicle is determined; the trajectory represented by the emergency braking trajectory is the trajectory that the autonomous vehicle is expected to travel within a preset time range when it cannot avoid colliding with other traffic participants.

7. The method as described in claim 1, characterized in that, The process of obtaining the target trajectory information corresponding to the autonomous vehicle using a pre-established neural network computing model specifically includes: Based on the trajectory coordinate transformation model, the latitude and longitude coordinates in the target trajectory information are determined, specifically including: converting the trajectory information in the Frenet coordinate system into latitude and longitude coordinates, and determining the latitude and longitude coordinate system as the latitude and longitude coordinates in the target trajectory information; the latitude and longitude coordinates include longitude, latitude, and heading angle.

8. The method as described in claim 2, characterized in that, Sending the target trajectory information to the autonomous vehicle so that the autonomous vehicle can follow the trajectory represented by the target trajectory information specifically includes: The autonomous vehicle acquires target trajectory information sent by the cloud control platform; Based on the vehicle feedback control algorithm model of the autonomous vehicle itself, and according to the deviation between the target trajectory information and the vehicle state information, control parameters for the vehicle actuators of the autonomous vehicle are determined; the vehicle actuators include a motor and a steering wheel; the control parameters include steering wheel angle, throttle opening or brake opening; The control parameters are used to control the vehicle actuators of the autonomous vehicle to follow the trajectory indicated by the target trajectory information.

9. An apparatus for planning the driving trajectory of an autonomous vehicle based on the method of any one of claims 1-8, characterized in that, include: The acquisition module is used by the cloud control platform to acquire vehicle status data sent by the autonomous vehicle; The acquisition module is further configured to acquire road environment data sent by roadside perception devices on the road where the autonomous vehicle is traveling; the road environment data includes driving data of other vehicles and pedestrian data within the driving area of ​​the autonomous vehicle. The calculation module is used to calculate the target trajectory information corresponding to the autonomous vehicle using a pre-established neural network model based on the vehicle status data and the road environment data; wherein, the neural network model is used to generate the target trajectory information of the autonomous vehicle; the trajectory represented by the target trajectory information is the trajectory that the autonomous vehicle is expected to travel within a preset time range; A sending module is used to send the target trajectory information to the autonomous vehicle so that the autonomous vehicle can follow the trajectory represented by the target trajectory information.

Images