A method, system, and autonomous vehicle for lane-changing decision-making.

Abstract

This invention belongs to the field of automotive technology and provides a method, system, and vehicle for lane-changing decision-making in autonomous vehicles. The technical solution involves preprocessing acquired road environment information; training a spatiotemporal feature extraction network based on the preprocessed road environment information to obtain a trained spatiotemporal feature extraction network; combining real-time perceived driving environment data and the trained feature extraction network to predict environmental perception results and lane-changing decision prediction results; fusing the environmental perception results and lane-changing decision prediction results to obtain a state representation; adjusting the optimal lane-changing strategy based on the state representation and action value function; performing path planning based on the adjusted optimal lane-changing strategy; and generating steering wheel angle signals and vehicle speed control signals based on the path planning results. This enables more accurate lane-changing decisions.

Description

Technical Field

[0001] This invention belongs to the field of vehicle-assisted driving, and particularly relates to a lane-changing decision-making method, system, and autonomous vehicle for autonomous driving. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Autonomous driving technology aims to enable vehicles to navigate and control autonomously through technologies such as computer vision and machine learning. Lane-changing decisions are a crucial aspect of autonomous driving, requiring vehicles to accurately perceive their surroundings, predict the behavior of other vehicles and pedestrians, and make appropriate lane-changing decisions accordingly.

[0004] However, road traffic environments are extremely complex and variable, including different road conditions, traffic flow, weather conditions, and the behavior of pedestrians and other vehicles. Traditional lane-changing decision-making methods, such as rule-based methods or simple machine learning methods, often fail to fully consider these complex factors and may therefore perform poorly in complex scenarios. Summary of the Invention

[0005] In view of this, the present invention aims to propose a lane-changing decision-making method for autonomous vehicles. By learning complex driving strategies and feature representations from a large amount of driving data, the method can achieve precise lane changing in the face of different complex situations, such as those involving obstacles or traffic congestion, by adjusting the steering wheel angle signal and vehicle speed control signal in real time. This reduces various risks encountered during lane changing and ensures safe and efficient driving.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A lane-changing decision-making method for autonomous vehicles includes the following steps:

[0008] The acquired road environment information is preprocessed;

[0009] The spatiotemporal feature extraction network is trained based on the preprocessed road environment information to obtain the trained spatiotemporal feature extraction network.

[0010] The environmental perception results and lane change decision prediction results are obtained by combining real-time perceived driving environment data and trained feature extraction network prediction.

[0011] The environmental perception results and lane change decision prediction results are fused to obtain a state representation, and the optimal lane change strategy is adjusted by combining the state representation and the action value function.

[0012] Path planning is performed based on the adjusted optimal lane-changing strategy, and steering wheel angle signals and vehicle speed control signals are generated based on the path planning results.

[0013] Furthermore, the construction process of the spatiotemporal feature extraction network includes:

[0014] The road environment information acquired from various sensors is input into a CNN network to obtain spatial features. Specifically, this includes: extracting feature data from different types of sensors by passing the input data through multiple convolutional layers; inputting real-time video frame data into the first convolutional layer to extract the spatial features of the image; inputting radar data into the second convolutional layer to extract the target distance and vehicle speed; and inputting LiDAR data into the third convolutional layer to extract the three-dimensional spatial structure and obstacle positions from the point cloud data.

[0015] The obtained spatial features and time series data are input into an RNN network to extract comprehensive spatiotemporal features.

[0016] Output lane-changing decision features based on comprehensive spatiotemporal characteristics.

[0017] Furthermore, the adjustment of the optimal lane-changing strategy by combining state representation and action value function includes:

[0018] Given a state representation, select an action 'a' based on the current action value function Q(s,a);

[0019] After performing action a, determine the new state s' obtained after performing action a, the optimal action a' selected in the new state, and the corresponding reward r, and calculate the action value of the optimal action a' and the new state s';

[0020] By maximizing the action value function Q(s,a), the optimal action, i.e. the optimal lane-changing strategy, is selected.

[0021] Furthermore, based on the path planning results, steering wheel angle signals and vehicle speed control signals are generated, including:

[0022] Select the target point as the control objective based on the path planning;

[0023] Based on the vehicle's current position, target point, and wheelbase, calculate the vehicle's tire steering angle and determine the steering wheel angle limit for the vehicle's tire steering angle.

[0024] Determine the target speed based on the suggested speed from the route planning and the current road conditions;

[0025] Calculate the throttle or brake signal based on the deviation between the current vehicle speed and the target speed.

[0026] Furthermore, the acquired road environment information is preprocessed, including cleaning and normalizing the collected raw data.

[0027] Furthermore, when calculating the optimal path, the path length and cost are considered for selection, and the final path planning result is obtained based on the weighted sum of the two.

[0028] Furthermore, the Adam optimizer is used when training the constructed spatiotemporal feature extraction network.

[0029] The second objective of this invention is to propose an autonomous vehicle lane-changing decision system. This system can learn complex driving strategies and feature representations from a large amount of driving data. In the face of different complex situations, such as those involving obstacles or traffic congestion, it can achieve precise lane changing by adjusting the steering wheel angle signal and vehicle speed control signal in real time, thereby reducing various risks encountered during lane changing and ensuring safe and efficient driving.

[0030] An autonomous vehicle lane-changing decision-making system includes:

[0031] The preprocessing module is used to preprocess the acquired road environment information;

[0032] The network training module is used to train the constructed spatiotemporal feature extraction network based on the preprocessed road environment information, so as to obtain the trained spatiotemporal feature extraction network.

[0033] The decision prediction module is used to combine real-time perceived driving environment data and trained feature extraction network prediction to obtain environmental perception results and lane change decision prediction results.

[0034] The decision adjustment module is used to fuse environmental perception results and lane change decision prediction results to obtain a state representation, and then adjust the optimal lane change strategy by combining the state representation and the action value function.

[0035] The control signal generation module is used for path planning based on the adjusted optimal lane-changing strategy, and generates steering wheel angle signals and vehicle speed control signals based on the path planning results.

[0036] The third objective of this invention is to propose an autonomous driving vehicle equipped with multiple sensors, including lidar sensors, cameras, and radar, which can acquire information such as road conditions, traffic flow, vehicle position, speed, and acceleration. When faced with complex environments, the vehicle generates lane-changing decision strategies by fusing the acquired multi-source data. In various complex situations, such as those involving obstacles or traffic congestion, the vehicle can achieve precise lane changes by adjusting the steering wheel angle signal and vehicle speed control signal in real time, reducing various risks encountered during lane changes and ensuring safe and efficient driving.

[0037] The third objective of this invention is to provide a program product.

[0038] A program product, which is a computer program product, includes a computer program that, when executed by a processor, implements the steps in the lane-changing decision method for an autonomous vehicle as described in the first objective.

[0039] Compared with the prior art, the beneficial effects of the present invention are:

[0040] This invention proposes a lane-changing decision-making method for autonomous vehicles. When faced with complex environments, the method generates a lane-changing decision-making strategy by fusing multi-source data. This strategy can achieve precise lane changing in various complex situations, such as those involving obstacles or traffic congestion, by adjusting the steering wheel angle signal and vehicle speed control signal in real time. This reduces various risks encountered during lane changing and ensures safe and efficient driving.

[0041] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0042] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0043] Figure 1 This is a flowchart of an autonomous vehicle lane-changing decision-making method according to an embodiment of the present invention;

[0044] Figure 2 This is a structural block diagram of an autonomous vehicle lane-changing decision system according to one embodiment of the present invention. Detailed Implementation

[0045] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0046] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0047] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0048] As mentioned in the background section, the road traffic environment is extremely complex and variable, including different road conditions, traffic flow, weather conditions, and the behavior of pedestrians and other vehicles. Traditional lane-changing decision-making methods, such as rule-based methods or simple machine learning methods, often fail to fully consider these complex factors and may therefore perform poorly in complex scenarios.

[0049] This invention proposes a lane-changing decision-making method for autonomous vehicles. When faced with complex environments, the method generates a lane-changing decision-making strategy by fusing multi-source data. This strategy can achieve precise lane changing in various complex situations, such as those involving obstacles or traffic congestion, by adjusting the steering wheel angle signal and vehicle speed control signal in real time. This reduces various risks encountered during lane changing and ensures safe and efficient driving.

[0050] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0051] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0052] Figure 1 This is a flowchart of a deep learning-based lane-changing decision-making method for autonomous driving according to an embodiment of the present invention.

[0053] like Figure 1 As shown, a deep learning-based lane-changing decision-making method for autonomous driving according to an embodiment of the present invention includes the following steps:

[0054] S101: Preprocess the acquired road environment information;

[0055] Among them, road environment information is collected in real time through a variety of sensors installed on autonomous vehicles;

[0056] Specifically, multiple sensors, including lidar sensors, cameras, and radar, acquire data that is road environment data collected by the corresponding lidar sensors, cameras, and radars; this information includes road conditions, traffic flow, vehicle position, speed, acceleration, etc.

[0057] In addition, considering the diversity and comprehensiveness of the data, relevant data is also obtained from other sources, such as traffic monitoring cameras and satellite positioning systems.

[0058] The preprocessing of the acquired road environment information includes cleaning and normalizing the collected raw data to remove noise, outliers and duplicate data, eliminate the dimensional differences between data from different sensors to give the data a uniform scale, and label the data to provide accurate input and output information for subsequent model training.

[0059] S102: The spatiotemporal feature extraction network is trained based on the preprocessed road environment information to obtain the trained spatiotemporal feature extraction network;

[0060] Specifically, the steps include the following:

[0061] S201. Input the road environment information acquired by each sensor into a CNN network to obtain spatial features, specifically including:

[0062] S2011. Input the road environment information acquired by each sensor into the input layer of the CNN network;

[0063] S2012: Extract feature data from different types of sensors by passing the input data through multiple convolutional layers;

[0064] Specifically, different convolution kernels are designed for different convolutional layers to extract corresponding features.

[0065] Specifically, the real-time video frame data acquired by the vehicle camera is input into the first convolutional layer, and the spatial features of the image are extracted through the first convolutional layer, including spatial features such as the image's edges, texture, and shape.

[0066] The radar data is input into the second convolutional layer to extract features such as target distance and vehicle speed from the radar data;

[0067] The LiDAR data is input into the third convolutional layer to extract features such as the three-dimensional spatial structure and obstacle location from the point cloud data.

[0068] For example, in the first layer: the kernel size is 3x3, the stride is 1, the padding is 1, and the number of output channels is 32.

[0069] The second layer has a kernel size of 3x3, a stride of 1, padding of 1, and 64 output channels.

[0070] The third layer has a kernel size of 3x3, a stride of 1, padding of 1, and 128 output channels.

[0071] S2013. The feature data of different types of sensors extracted are subjected to nonlinear transformation by activation functions and downsampled by pooling layers to reduce the spatial dimension of the data and retain important spatial features.

[0072] For example, using a max pooling layer, the pooling window size is 2x2, and the stride is 2. Pooling layers can reduce the size of feature maps and lower computational complexity.

[0073] S2014. After flattening the outputs of each convolutional layer and pooling layer, spatial features are fused through a fully connected layer to obtain fused spatial features.

[0074] Specifically, the features extracted by the convolutional layers are fused and a preliminary feature vector is output.

[0075] The structure of the fully connected layer includes: Layer 1: 512 neurons; Layer 2: 256 neurons.

[0076] S2014. Input the spatial features obtained from S2013 into the RNN network, and simultaneously input time series data such as vehicle speed, acceleration, and steering angle into the RNN network.

[0077] S2015. Spatial features and time series data are processed through mathematical operations such as addition or splicing, and then nonlinear transformations are applied before inputting them into the loop layer to extract comprehensive spatiotemporal features that have an important impact on lane-changing decisions.

[0078] Specifically, the recurrent layer uses recurrent units such as LSTM or GRU to capture long-term dependencies in time series data;

[0079] S2016 Finally, the comprehensive spatiotemporal features that have a significant impact on lane-changing decisions are output as lane-changing decision features through a fully connected layer.

[0080] Among them, the lane-changing decision features include whether to change lanes (by outputting the probability through the softmax function and making a judgment), the direction of lane changing (by outputting the probability through the softmax function and making a judgment), and the timing of lane changing (by outputting a specific value or timestamp).

[0081] The output layer can set the number of output nodes according to task requirements, such as lane change decision results, steering wheel angle, etc.

[0082] S202. Train the constructed spatiotemporal feature extraction network to obtain the trained spatiotemporal feature extraction network.

[0083] S2021, Selection of training algorithm;

[0084] Among them, the training algorithm chosen is the Adam optimizer, which generally performs well because it combines momentum and adaptive learning rate methods.

[0085] Loss function: Cross-entropy loss is used;

[0086] For example, mean squared error loss can be used for steering wheel angle prediction.

[0087] S2022, Training process: Forward propagation: Input data is fed into the network to obtain the output; Loss calculation: The loss is calculated based on the output and the true label; Backward propagation: The gradient is calculated and the weights are updated.

[0088] S2023, Parameter Adjustment: Learning Rate: The initial learning rate is usually set to a small value (such as 0.001) and gradually decreased during training; Batch Size: Select an appropriate batch size based on hardware resources, such as 32, 64, or 128; Number of Iterations: Set the total number of iterations (Epochs), usually starting with 10 and gradually increasing until the model converges; Regularization: Use L1 or L2 regularization to prevent overfitting.

[0089] S103: Input the real-time perceived driving environment data into the trained feature extraction network to predict the best time and path for lane changing;

[0090] Specifically, the steps include the following:

[0091] S301, Real-time perception of driving environment data acquisition module, including driving environment data acquired by cameras, radar, and lidar;

[0092] S302. Input the real-time perceived driving environment data into the CNN network, and determine the current perception result based on the output features, such as the vehicle's position, speed, obstacle information, etc.

[0093] For example, it can be determined whether the current traffic is congested; specifically, determining whether the current road is a congested section involves:

[0094]

[0095] Where Congestion represents the traffic congestion state; ρ represents the traffic density; v represents the average speed; and ρ th v is the traffic density threshold. th The average speed threshold;

[0096] S303. By combining time series data and RNN network, the lane-changing decision prediction results are obtained;

[0097] The RNN unit update rule is as follows:

[0098] h t =f(Wxh x t +W hh h t-1 +b h ),

[0099] Among them, h t Let x be the hidden state at the current moment. t h is the input at the current moment. t-1 It's the hidden state from the previous moment, W xh and W hh It is the weight matrix, b h f(·) is the bias term, and f(·) is the activation function, usually ReLU or Tanh.

[0100] S104. Integrate the perception results and lane change decision prediction results to obtain a unified state representation s, and combine the state representation s with the action value function to select the best lane change strategy.

[0101] Specifically, the steps include the following:

[0102] S401. First, the perception results (such as vehicle position, speed, obstacle information, etc.) and the prediction information (such as possible traffic flow and road condition changes in the future) are fused into a unified state representation s. This state representation s is the key input in the reinforcement learning algorithm, and it contains all the information needed for decision-making.

[0103] S402, Action Selection: Given a state s, select an action a based on the current action value function Q(s,a). This action a may represent changing lanes to the left lane, changing lanes to the right lane, or keeping the current lane, etc.

[0104] S403. Reward Calculation: After performing action a, a reward r is calculated based on the result. This reward r reflects the immediate benefits gained after performing the action (such as reducing traffic delays, improving safety, etc.).

[0105] S404, Action Value Function Update: Finally, the action value function Q(s,a) is updated using the update formula in reinforcement learning; the best action is selected by maximizing the action value function Q(s,a).

[0106]

[0107] Q(s,a)←Q(s,a)+α[R(s,a)+γmaxQ(s′,a′)-Q(s,a)],

[0108] Where α is the learning rate, γ is the discount factor, and Q(s′,a′) is the action value of the optimal action a' and the new state s'.

[0109] This update formula typically includes the current action value Q(s,a), the new state s' obtained after performing the action, the optimal action a' chosen in the new state, and the corresponding reward r. Perception results and prediction information indirectly participate in the update process of the action value function Q(s,a) by influencing the state representation s and reward r.

[0110] S105: Based on the optimal timing and path for lane changing, perform path planning and generate control signals for vehicle operation based on the path planning results;

[0111] Specifically, the steps include the following:

[0112] S501. Calculate the optimal path from the current location to the target lane;

[0113] In this embodiment, when calculating the optimal path, the path length and cost are considered. Specifically, each result can be assigned a weight according to the actual situation, and the final path planning result is obtained based on the weighted sum of the two.

[0114] Specifically, the A* algorithm can be used to find the shortest path, and the cost of the path is evaluated according to the cost function C, which is expressed as:

[0115]

[0116] Among them, c i It is the cost of each step on the path, and n is the total number of steps the path takes.

[0117] S502. Based on the path planning results, generate control signals to control vehicle operation and control vehicle operation.

[0118] In this embodiment, the control signal includes a steering wheel angle signal and a vehicle speed signal;

[0119] Specifically, the steps include the following:

[0120] S5021. Path planning includes a series of waypoints, each containing location coordinates (x, y), speed suggestions, curvature, and other information. Based on the current vehicle position, a target point is selected as the control objective.

[0121] S5022. Based on the vehicle's current position, target point, and wheelbase, calculate the vehicle's tire steering angle and determine the steering wheel angle limit for the vehicle's tire steering angle.

[0122] The formula for calculating the steering angle is:

[0123]

[0124] Where θ is the vehicle's steering angle, L is the vehicle's wheelbase, and ω is the angle between the vehicle's current heading and the target point's direction. d It is the distance from the vehicle to the target point.

[0125] S5023. Determine the target speed based on the suggested speed of the route planning and the current road conditions (such as curvature, obstacles, etc.);

[0126] S5024. Calculate the throttle or brake signal based on the deviation between the current vehicle speed and the target speed;

[0127]

[0128] Where u(t) is the control output of the throttle or brake, e(t) is the speed error signal between the target speed and the current speed, and K p K i K d These are PID parameters.

[0129] Through the above steps, the autonomous driving system can adjust its lane-changing strategy in a timely manner when it detects obstacles or traffic congestion ahead, ensuring safe and efficient driving.

[0130] Figure 2 This is a structural block diagram of an autonomous vehicle lane-changing decision-making system according to an embodiment of the present invention. Figure 2 As shown, an autonomous vehicle lane-changing decision system according to an embodiment of the present invention includes: a preprocessing module, a network training module, a decision prediction module, a decision adjustment module, and a control signal generation module.

[0131] The segmentation module is used to preprocess the acquired road environment information;

[0132] The network training module is used to train the constructed spatiotemporal feature extraction network based on the preprocessed road environment information, so as to obtain the trained spatiotemporal feature extraction network.

[0133] The decision prediction module is used to combine real-time perceived driving environment data and trained feature extraction network prediction to obtain environmental perception results and lane change decision prediction results.

[0134] The decision adjustment module is used to fuse environmental perception results and lane change decision prediction results to obtain a state representation, and then adjust the optimal lane change strategy by combining the state representation and the action value function.

[0135] The control signal generation module is used for path planning based on the adjusted optimal lane-changing strategy, and generates steering wheel angle signals and vehicle speed control signals based on the path planning results.

[0136] It should be noted that the specific implementation of the autonomous vehicle lane-changing decision system in this embodiment of the invention is similar to the specific implementation of the autonomous vehicle lane-changing decision method in this embodiment of the invention. Please refer to the description in the method section for details. To reduce redundancy, it will not be repeated here.

[0137] Furthermore, embodiments of the present invention disclose an autonomous vehicle equipped with an autonomous vehicle lane-changing decision system as described in the above embodiments. The vehicle is equipped with various sensors, including lidar sensors, cameras, and radar, which can acquire information such as road conditions, traffic flow, vehicle position, speed, and acceleration. When faced with complex environments, the system generates a lane-changing decision strategy by fusing the acquired multi-source data. This allows for precise lane changes in various complex situations, such as those involving obstacles or traffic congestion, by adjusting the steering wheel angle signal and vehicle speed control signal in real time. This reduces various risks encountered during lane changes and ensures safe and efficient driving.

[0138] Furthermore, other components and functions of the vehicle according to embodiments of the present invention are known to those skilled in the art, and will not be described in detail here to reduce redundancy.

[0139] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0140] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A lane-changing decision-making method for autonomous vehicles, characterized in that, Includes the following steps: The acquired road environment information is preprocessed; The spatiotemporal feature extraction network is trained based on the preprocessed road environment information to obtain the trained spatiotemporal feature extraction network. The environmental perception results and lane change decision prediction results are obtained by combining real-time perceived driving environment data and trained feature extraction network prediction. The environmental perception results and lane change decision prediction results are fused to obtain a state representation, and the optimal lane change strategy is adjusted by combining the state representation and the action value function. Based on the adjusted optimal lane-changing strategy, path planning is performed, and steering wheel angle signals and vehicle speed control signals are generated according to the path planning results. The construction process of the spatiotemporal feature extraction network includes: The road environment information acquired from various sensors is input into a CNN network to obtain spatial features. Specifically, this includes: extracting feature data from different types of sensors by passing the input data through multiple convolutional layers; inputting real-time video frame data into the first convolutional layer to extract the spatial features of the image; inputting radar data into the second convolutional layer to extract the target distance and vehicle speed; and inputting LiDAR data into the third convolutional layer to extract the three-dimensional spatial structure and obstacle positions from the point cloud data. The obtained spatial features and time series data are input into an RNN network to extract comprehensive spatiotemporal features. Output lane-changing decision features based on comprehensive spatiotemporal characteristics; The method of adjusting the optimal lane-changing strategy by combining state representation and action value function includes: Given the state representation, based on the current action value function Choose an action ; Execute action Then, determine the new state obtained after performing the action. The optimal action to choose in the new state and corresponding rewards r Calculate the optimal action and new status The value of the action; By maximizing the action value function Choose the best action, that is, the best lane-changing strategy.

2. The lane-changing decision-making method for autonomous vehicles as described in claim 1, characterized in that, Based on the path planning results, steering wheel angle signals and vehicle speed control signals are generated, including: Select the target point as the control objective based on the path planning; Based on the vehicle's current position, target point, and wheelbase, calculate the vehicle's tire steering angle and determine the steering wheel angle limit for the vehicle's tire steering angle. Determine the target speed based on the suggested speed from the route planning and the current road conditions; Calculate the throttle or brake signal based on the deviation between the current vehicle speed and the target speed.

3. The lane-changing decision-making method for autonomous vehicles as described in claim 1, characterized in that, The acquired road environment information is preprocessed, including cleaning and normalizing the collected raw data.

4. The lane-changing decision-making method for autonomous vehicles as described in claim 1, characterized in that, When calculating the optimal path, the path length and cost are considered, and the final path planning result is obtained based on the weighted sum of the two.

5. The lane-changing decision-making method for autonomous vehicles as described in claim 1, characterized in that, The Adam optimizer is used when training the constructed spatiotemporal feature extraction network.

6. A lane-changing decision-making system for autonomous vehicles, characterized in that, include: The preprocessing module is used to preprocess the acquired road environment information; A network training module is used to train a constructed spatiotemporal feature extraction network based on preprocessed road environment information, resulting in a trained spatiotemporal feature extraction network; wherein, the construction process of the spatiotemporal feature extraction network includes: The road environment information acquired from various sensors is input into a CNN network to obtain spatial features. Specifically, this includes: extracting feature data from different types of sensors by passing the input data through multiple convolutional layers; inputting real-time video frame data into the first convolutional layer to extract the spatial features of the image; inputting radar data into the second convolutional layer to extract the target distance and vehicle speed; and inputting LiDAR data into the third convolutional layer to extract the three-dimensional spatial structure and obstacle positions from the point cloud data. The obtained spatial features and time series data are input into an RNN network to extract comprehensive spatiotemporal features. Output lane-changing decision features based on comprehensive spatiotemporal characteristics; The decision prediction module is used to combine real-time perceived driving environment data and trained feature extraction network prediction to obtain environmental perception results and lane change decision prediction results. The decision adjustment module is used to fuse environmental perception results and lane change decision prediction results to obtain a state representation, and then adjust the optimal lane change strategy by combining the state representation and the action value function. The method of adjusting the optimal lane-changing strategy by combining state representation and action value function includes: Given the state representation, based on the current action value function Choose an action ; Execute action Then, determine the new state obtained after performing the action. The optimal action to choose in the new state and corresponding rewards r Calculate the optimal action and new status The value of the action; By maximizing the action value function Choose the best action, that is, the best lane-changing strategy; The control signal generation module is used for path planning based on the adjusted optimal lane-changing strategy, and generates steering wheel angle signals and vehicle speed control signals based on the path planning results.

7. An autonomous driving vehicle, characterized in that, An autonomous vehicle lane-changing decision system as described in claim 6 is provided.

8. A program product, said program product being a computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps in the lane-changing decision method for an autonomous vehicle as described in any one of claims 1-5.

Images