Motion state planning method and device based on time domain convolutional neural network

Abstract

The invention relates to the technical field of vehicles, in particular to a motion state planning method and device based on a time domain convolutional neural network, and the method comprises the steps: obtaining the motion state information of a current vehicle at a current moment and the motion state information of at least one historical moment adjacent to the current moment; inputting the motion state information of the current moment and the motion state information of at least one historical moment into a preset motion state planning model for prediction to obtain predicted motion state information of the current vehicle; and in combination with a preset loss function, according to the predicted motion state information of the current vehicle, generating a lane changing control instruction of the vehicle. Therefore, the problems of low historical data utilization rate, insufficient dynamic scene adaptability and high calculation complexity in the prior art are solved, the motion state of the next position of the vehicle is accurately predicted, the real-time performance, the accuracy and the scene adaptability of speed planning are effectively improved, and stable and comfortable driving experience is provided for a user.

Description

Technical Field

[0001] This application relates to the field of vehicle technology, and in particular to a motion state planning method and apparatus based on a temporal convolutional neural network. Background Technology

[0002] When an autonomous vehicle (SV) detects that a vehicle ahead is slowing down or moving at a low speed, the system triggers a lane-changing decision. It plans and executes a lane-changing trajectory to avoid obstruction and ensure smooth driving. Before executing the lane-changing operation, the autonomous vehicle needs to complete the lane-changing trajectory planning, employing a path-velocity decoupling strategy for step-by-step processing: first, a geometric path is generated based on static environmental constraints; then, speed and time allocation are optimized according to dynamic conditions. This method, by separating path planning (spatial dimension) and speed planning (time dimension), significantly reduces computational complexity while ensuring trajectory feasibility and improving the system's real-time response capability and modular design flexibility.

[0003] Current speed planning methods are mainly divided into two categories: rule-based methods and optimization-based methods. Rule-based methods include trapezoidal speed curve methods, S-curve speed curve methods, and piecewise constant speed methods. The trapezoidal speed curve method uses a three-stage acceleration-constant speed-deceleration approach to achieve basic speed control, suitable for structured road scenarios. The S-curve speed curve method introduces jerk constraints on top of the trapezoidal curve. The piecewise constant speed method implements a conservative strategy through path segmentation and fixed speed allocation (such as deceleration on curves and acceleration on straightaways), but its applicability is limited to known environments. Optimization-based methods mainly include Quadratic Programming (QP), Model Predictive Control (MPC), and Dynamic Programming (DP). Quadratic programming optimizes acceleration, Jerk, and other indices under path constraints to generate smooth speed curves, but its adaptability in dynamic environments is poor. MPC and DP methods, on the other hand, have high computational costs, are time-consuming, and have poor real-time performance.

[0004] Therefore, while existing methods each have their advantages, they still have certain limitations. On the one hand, these methods are inefficient in utilizing historical driving data, making it difficult to fully explore empirical patterns; on the other hand, they have high computational complexity, and their adaptability in complex dynamic scenarios still needs further improvement, which urgently needs to be addressed. Summary of the Invention

[0005] This application provides a motion state planning method and apparatus based on Temporal Convolutional Network (TCN) to solve the problems of low utilization of historical data, insufficient adaptability to dynamic scenes, and high computational complexity in the prior art. It achieves accurate prediction of the motion state of the vehicle at the next position, effectively improves the real-time performance, accuracy, and scene adaptability of speed planning, and provides users with a smooth and comfortable driving experience.

[0006] The first aspect of this application provides a motion state planning method based on a temporal convolutional neural network, including the following steps:

[0007] Obtain the motion state information of the current vehicle at the current moment and the motion state information of at least one historical moment adjacent to the current moment;

[0008] The motion state information at the current moment and the motion state information at at least one historical moment are input into a preset motion state planning model for prediction to obtain the predicted motion state information of the current vehicle.

[0009] Based on a preset loss function and the predicted motion state information of the current vehicle, a lane-changing control command is generated for the vehicle.

[0010] According to one embodiment of this application, the preset loss function is determined by a penalty function for the impact of surrounding vehicle interactions, a robust loss function, and a user comfort penalty function.

[0011] According to one embodiment of this application, the function for the penalty term related to the interaction with surrounding vehicles is:

[0012] P F(TV) =k1·f(TV1)+k2·f(TV2)+k3·f(TV3);

[0013] Among them, P F(TV) Let f(TV1) be the function of the penalty term for the interaction influence of surrounding vehicles, k1 be the influence coefficient of f(TV1), f(TV1) be the function of the current vehicle being affected by vehicle TV1, k2 be the influence coefficient of f(TV2), f(TV2) be the function of the current vehicle being affected by vehicle TV2, k3 be the influence coefficient of f(TV3), and f(TV3) be the function of the current vehicle being affected by vehicle TV3.

[0014] The user comfort penalty function is:

[0015]

[0016] Among them, P P (M T+1 ) is a function for user comfort penalty terms. q1 is the weight of velocity V, q2 is the weight of acceleration a, and q3 is the weight of heading angle θ. v th For the threshold value of velocity V, a th For the threshold of acceleration a, θ th M is the threshold value for the heading angle θ. T+1 For the current vehicle at P T+1 List of status information for point M T For the current vehicle at P T The list of status information for points, where ΔT represents the current vehicle's position from point P. T Point moves to P T+1 Time taken.

[0017] According to one embodiment of this application, the preset loss function is:

[0018]

[0019] Among them, Loss P The default loss function is... Let P be the robust loss function, λ1 be the weighting factor for the penalty term considering passenger comfort, and P be the weighting factor for the robust loss function. F(TV) Let λ be the penalty term function for the interaction effects of surrounding vehicles, and λ2 be the weighting factor for the penalty term considering the interaction effects of surrounding vehicles. To account for the penalty factor that influences interactions with surrounding vehicles, M represents the actual value of the vehicle's state information list at the predicted location point. This provides the predicted motion state information for the current vehicle.

[0020] According to one embodiment of this application, the motion state information of the current vehicle at the current moment includes at least one of the current vehicle speed, the current acceleration, and the current heading angle.

[0021] According to the motion state planning method based on a temporal convolutional neural network provided in this application, the motion state information of the current vehicle at the current moment and the motion state information at at least one historical moment are input into a preset motion state planning model for prediction to obtain the predicted motion state information of the current vehicle; combined with a preset loss function, lane-changing control commands for the vehicle are generated. This solves the problems of low utilization of historical data, insufficient adaptability to dynamic scenarios, and high computational complexity in the prior art, achieving accurate prediction of the vehicle's next position motion state, effectively improving the real-time performance, accuracy, and scene adaptability of speed planning, and providing users with a smooth and comfortable driving experience.

[0022] A second aspect of this application provides a motion state planning device based on a temporal convolutional neural network, comprising:

[0023] The acquisition module is used to acquire the motion state information of the current vehicle at the current moment and the motion state information of at least one historical moment adjacent to the current moment;

[0024] The prediction module is used to input the motion state information at the current moment and the motion state information at at least one historical moment into a preset motion state planning model for prediction, so as to obtain the predicted motion state information of the current vehicle.

[0025] The planning module is used to generate lane-changing control commands for the vehicle based on the predicted motion state information of the current vehicle, combined with a preset loss function.

[0026] According to one embodiment of this application, the preset loss function is determined by a penalty function for the impact of surrounding vehicle interactions, a robust loss function, and a user comfort penalty function.

[0027] According to one embodiment of this application, the function for the penalty term related to the interaction with surrounding vehicles is:

[0028] P F(TV) =k1·f(TV1)+k2·f(TV2)+k3·f(TV3);

[0029] Among them, P F(TV) Let f(TV1) be the function of the penalty term for the interaction influence of surrounding vehicles, k1 be the influence coefficient of f(TV1), f(TV1) be the function of the current vehicle being affected by vehicle TV1, k2 be the influence coefficient of f(TV2), f(TV2) be the function of the current vehicle being affected by vehicle TV2, k3 be the influence coefficient of f(TV3), and f(TV3) be the function of the current vehicle being affected by vehicle TV3.

[0030] The user comfort penalty function is:

[0031]

[0032] Among them, P P (M T+1 ) is a function for user comfort penalty terms. q1 is the weight of velocity V, q2 is the weight of acceleration a, and q3 is the weight of heading angle θ. v th For the threshold value of velocity V, a th For the threshold of acceleration a, θ th M is the threshold value for the heading angle θ. T+1 For the current vehicle at P T+1 List of status information for point M T For the current vehicle at P T The list of status information for points, where ΔT represents the current vehicle's position from point P. TPoint moves to P T+1 Time taken.

[0033] According to one embodiment of this application, the preset loss function is:

[0034]

[0035] Among them, Loss P The default loss function is... Let P be the robust loss function, λ1 be the weighting factor for the penalty term considering passenger comfort, and P be the weighting factor for the robust loss function. F(TV) Let λ be the penalty term function for the interaction effects of surrounding vehicles, and λ2 be the weighting factor for the penalty term considering the interaction effects of surrounding vehicles. To account for the penalty factor that influences interactions with surrounding vehicles, M represents the actual value of the vehicle's state information list at the predicted location point. This provides the predicted motion state information for the current vehicle.

[0036] According to one embodiment of this application, the motion state information of the current vehicle at the current moment includes at least one of the current vehicle speed, the current acceleration, and the current heading angle.

[0037] According to the motion state planning device based on a temporal convolutional neural network provided in this application embodiment, the motion state information of the current vehicle at the current moment and the motion state information at at least one historical moment are input into a preset motion state planning model for prediction to obtain the predicted motion state information of the current vehicle; combined with a preset loss function, lane-changing control commands for the vehicle are generated. This solves the problems of low utilization of historical data, insufficient adaptability to dynamic scenarios, and high computational complexity in the prior art, achieving accurate prediction of the vehicle's next position motion state, effectively improving the real-time performance, accuracy, and scene adaptability of speed planning, and providing users with a smooth and comfortable driving experience.

[0038] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the program to implement the motion state planning method based on a temporal convolutional neural network as described in the above embodiments.

[0039] A fourth aspect of this application provides a computer-readable storage medium storing computer instructions for causing the computer to execute the motion state planning method based on a temporal convolutional neural network as described in the above embodiments.

[0040] Additional aspects and advantages of this application 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 this application. Attached Figure Description

[0041] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0042] Figure 1 This is a flowchart of a motion state planning method based on a temporal convolutional neural network according to an embodiment of this application;

[0043] Figure 2 This is a schematic diagram of a lane-changing scenario according to an embodiment of this application;

[0044] Figure 3 This is an architecture diagram of a preset motion state planning model according to an embodiment of this application;

[0045] Figure 4 This is a block diagram of a motion state planning device based on a temporal convolutional neural network according to an embodiment of this application;

[0046] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0047] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0048] The following description, with reference to the accompanying drawings, describes a motion state planning method and apparatus based on a temporal convolutional neural network according to embodiments of this application. Addressing the problems of low utilization of historical data, insufficient adaptability to dynamic scenes, and high computational complexity mentioned in the background art, this application provides a motion state planning method based on a temporal convolutional neural network. In this method, the motion state information of the current vehicle at the current moment and the motion state information at at least one historical moment are input into a preset motion state planning model for prediction, obtaining the predicted motion state information of the current vehicle; combined with a preset loss function, a lane-changing control command for the vehicle is generated. This solves the problems of low utilization of historical data, insufficient adaptability to dynamic scenes, and high computational complexity in the background art, achieving accurate prediction of the vehicle's next position motion state, effectively improving the real-time performance, accuracy, and scene adaptability of speed planning, and providing users with a smooth and comfortable driving experience.

[0049] Specifically, Figure 1 This is a flowchart illustrating a motion state planning method based on a temporal convolutional neural network, provided in an embodiment of this application.

[0050] like Figure 1 As shown, the motion state planning method based on a temporal convolutional neural network includes the following steps:

[0051] In step S101, the motion state information of the current vehicle at the current moment and the motion state information of at least one historical moment adjacent to the current moment are obtained.

[0052] Furthermore, in some embodiments, the motion state information of the current vehicle at the current moment includes at least one of the current vehicle speed, the current acceleration, and the current heading angle.

[0053] Specifically, such as Figure 2 The diagram illustrates a lane-changing scenario. There are three lanes on the road, labeled L1, L2, and L3. In lane L1, there are two vehicles traveling in the same direction, TV1 (test vehicle 1) and TV2 (test vehicle 2), with speeds V... TV1 and V TV2 In lane L2, there is a vehicle SV (the tested vehicle) changing lanes. Its position at time T is P. T The velocity can be decomposed into V x (Longitudinal velocity) and V y (Lateral velocity), with a heading angle of α. The positions of the tested vehicle SV at previous times Tm, ..., T-1 are P, ... T-m ,…,P T-1 and at a speed V T Driving. The vehicle's SV predicts its position as P at future times T+1,…,T+n. T+1 ,…,P T+n In lane L2, ahead of the test vehicle SV is vehicle TV3 (a faulty or slow-moving test vehicle), traveling at a speed of V. TV3 The distance between the tested vehicle SV and TV3 is d. Tv3 Along the road direction, the distances between the tested vehicle SV and TV1 and TV2 are d respectively. TV1 d TV2 .

[0054] The speed planning method in this application is based on the path-velocity decoupling framework of trajectory planning and assumes that the search for the next path location point has been completed. This speed planning method is mainly aimed at the situation where an autonomous vehicle cuts into an adjacent lane when changing lanes. During the speed planning process, considering that there should be no sudden changes in adjacent moments before and after the vehicle, the planned speed and acceleration / deceleration should take into account passenger comfort. Therefore, the speed and acceleration of the next moment are related to the current motion state (speed, acceleration) and the motion state of the location points at several adjacent moments in the past.

[0055] like Figure 2 As shown, P T Point P represents the current location of the tested vehicle SV at the current moment. T+1 The point represents the next location of the current vehicle SV. The current vehicle SV is determined by P. T-1 Point moves to P T+n The point completes the lane change and merges into the adjacent lane. This embodiment of the application follows a path-velocity decoupling design method based on trajectory planning. The current vehicle SV completes the P... T+n After planning the location of the points, then start P. T+n Motion state planning for a point. The current vehicle SV moves from the planned lane change start point to the lane change completion point P. T+n During the process, the motion state planning is affected by factors such as the vehicle in front of the adjacent lane (TestVehicle2, TV2), the vehicle behind the adjacent lane (TestVehicle1, TV1), and the vehicle in front of the current lane (TestVehicle3, TV3).

[0056] P T+1 Factors affecting the velocity and acceleration planning of a point include: historical point P T-N+1 ,…,P T-1 motion state, P T Point motion state, P T The vehicle's heading angle, the position and movement of traffic participants ahead in this lane, and the position and movement of traffic participants before and after changing lanes.

[0057] The current vehicle SV is in P T The state information of the point includes: velocity V T acceleration a T , heading angle θ T The current vehicle SV is in P. T Point status information list M T It is expressed as follows:

[0058] M T =[V T ,a T ,θ T (1)(3) Vehicle SV is in P TList of historical and current status information at a given time

[0059] Based on similar methods, a list M of the current vehicle SV's state information at historical location points can be generated. T-i :

[0060] M T-i =[V T-i ,a T-i ,θ T-i ], i∈(0,N-1) (2)

[0061] Among them, V T-i Let a be the speed of the current vehicle SV at time Ti. T-i Let θ be the acceleration of the current vehicle SV at time Ti. T-i Let be the heading angle of the current vehicle SV at time Ti.

[0062] Furthermore, select P at the current time T The N-1 adjacent points of point P are used as historical experience information, and are compared with the current point P. T The status information consists of the input that the current vehicle SV is in P. T Point Information Time Series ISq T :

[0063]

[0064] That is, the preset motion state planning model MotPreTCN in P T Point prediction P T+1 The input information for the motion state of a point is represented as follows:

[0065]

[0066] In step S102, the motion state information at the current moment and the motion state information at at least one historical moment are input into a preset motion state planning model for prediction to obtain the predicted motion state information of the current vehicle.

[0067] First, we will introduce the architecture of the pre-defined motion state planning model proposed in this application.

[0068] Specifically, such as Figure 3 As shown, the preset motion state planning model MotPreTCN in this application embodiment mainly includes a convolutional layer, a TCN (Temporal Convolutional Network) module, and an MLP (Multilayer Perceptron) module.

[0069] The convolutional network layer consists of one-dimensional convolutions (Conv1d), primarily used to adjust the size of the output time-series data for easier processing by subsequent modules. The TCN module mainly comprises four two-dimensional dilated convolutional layers (Layer 1 to Layer 4), each consisting of a two-dimensional dilated convolution and an activation function layer, used to extract feature information from the input time series. The MLP module mainly comprises four fully connected layers, used to predict the motion state information of the next location point based on the features extracted by the TCN.

[0070] The prediction process of the preset motion state planning model MotPreTCN in this application embodiment is described below.

[0071] First, determine that the current vehicle is in position P. T Point Information Time Series ISq T This information is a time series ISq T By P T M is a list of state information of N-1 adjacent points. T-i ,i∈(0,N-1) and P T Point status information list M T The information time series ISq is constructed. T Input the preset motion state planning model MotPreTCN.

[0072] Secondly, with P T M is a list of state information of N-1 adjacent points. T-i The state information of each point in the network is processed by a one-dimensional convolution after passing through a convolutional network layer. Then, all the processed M... T-i Status information is integrated into two-dimensional data;

[0073] Furthermore, the TCN module extracts temporal and spatial feature information from the two-dimensional data processed by the convolutional network layers;

[0074] Finally, based on the feature information extracted by the TCN module, the MLP module predicts the motion state information of the next location point.

[0075] In step S103, a lane-changing control command for the vehicle is generated based on the predicted motion state information of the current vehicle, combined with a preset loss function.

[0076] Furthermore, in some embodiments, the preset loss function is determined by a penalty function for the impact of surrounding vehicle interactions, a robust loss function, and a user comfort penalty function.

[0077] Specifically, during the prediction process of the preset motion state planning model, the interaction information in the lane-changing scenario will have an impact, and this impact is embedded into the model parameter expression during the model training phase. Based on this, the embodiments of this application construct corresponding penalty terms according to the impact characteristics of the interaction information in the lane-changing scenario when setting the loss function.

[0078] Furthermore, such as Figure 2 As shown, during the current vehicle's lane change process, the vehicle is affected by the motion states of test vehicle TV1, test vehicle TV2, and the vehicle ahead TV3, as well as their relative distances to the current vehicle's lane. The safe distance between the moving vehicles is set to d. s The current vehicle SV must not obstruct the movement of vehicles in the adjacent lane when merging into the adjacent lane. Test vehicle TV1, test vehicle TV2, and the vehicle ahead TV3 are set to move at constant speeds of V. TV1 V TV2 and V TV3 The entire lane-changing process begins when the front of the vehicle touches the lane line and ends when the rear of the vehicle is fully inside the lane.

[0079] First, consider the interactive impact of the test vehicle TV1.

[0080] Current vehicle SV motion P T+1 When the point moves, the current motion state information of the vehicle SV is M. T+1 =[V T+1 ,a T+1 ,θ T+1 ].

[0081] In this application embodiment, an identification factor μ is used to indicate whether the vehicle has completed a lane change. If μ < 0, it means that the vehicle has not yet entered the adjacent lane; if μ = 0, it means that the vehicle is driving on the lane line and is entering the adjacent lane; if μ > 0, it means that the vehicle has completely entered the adjacent lane and completed the lane change.

[0082] To maintain a sufficient safe distance from the test vehicle TV1, the collision time was set to be greater than the first preset threshold TTC. th1 At that time, the current vehicle SV and the test vehicle TV1 will not collide. Considering that the current vehicle SV is at P T+1 The velocity and acceleration of the point, and the relative distance d between the current vehicle SV and the test vehicle TV1. sv_TV1 Therefore, the following constraints need to be applied to the movement of SV when it enters an adjacent lane:

[0083]

[0084] Among them, V T+1 For the current vehicle SV in P T+1The velocity of the point, V TV1 To test the speed of vehicle TV1, a T+1 For the current vehicle SV in P T+1 The acceleration of a point, d sv_TV1 The relative distance between the current vehicle SV and the test vehicle TV1, TTC th1 This is the first preset threshold.

[0085] Therefore, we can obtain:

[0086]

[0087] It can be inferred that the current vehicle's SV is in P T+1 The motion state of a point is affected by the vehicle TV1 function f(TV1) as follows:

[0088]

[0089] f(TV1) should approach zero before the lane change is completed to ensure that the test vehicle TV1 does not collide with the current vehicle SV.

[0090] Secondly, consider the interactive impact of the test vehicle TV2.

[0091] This application embodiment employs the same inference process as described above, considering the interaction effects of test vehicle TV1, to analyze the impact of test vehicle TV2 on the current vehicle SV. Taking into account that the current vehicle SV maintains a sufficient safe distance from test vehicle TV2 after changing lanes, the collision time is set to be greater than a second preset threshold TTC. th2 At that time, the current vehicle SV and the test vehicle TV2 will not collide. The current vehicle SV is at P T+1 The velocity and acceleration of the point, and the relative distance d between the current vehicle SV and the test vehicle TV2. sv_TV2 Therefore, when the current vehicle SV enters the adjacent lane, the following constraints need to be applied to its movement:

[0092]

[0093] Among them, V T+1 For the current vehicle SV in P T+1 The velocity of the point, V TV2 To test the speed of vehicle TV2, a T+1 For the current vehicle SV in P T+1 The acceleration of a point, d sv_TV2 The relative distance between the current vehicle SV and the test vehicle TV2, TTC th2 This is the second preset threshold.

[0094] It can be inferred that SV is in P T+1The motion state of a point is affected by the vehicle TV2 function f(TV2) as follows:

[0095]

[0096] Furthermore, the interactive effects of vehicle TV3 are considered.

[0097] Specifically, the impact of the vehicle ahead (TV3) on the current vehicle (SV) is analyzed. Before changing lanes, the current vehicle (SV) should maintain a sufficient safe distance from the vehicle ahead (TV3), and the collision time should be set to be greater than the third preset threshold (TTC). th3 At that time, the current vehicle SV will not collide with the vehicle TV3 in front. The current vehicle SV is in P T+1 The velocity and acceleration of the point, and the relative distance d between the current vehicle SV and TV3. sv_TV3 Therefore, when the current vehicle SV has not entered the adjacent lane, the following constraints need to be applied to its movement:

[0098]

[0099] Among them, V T+1 For the current vehicle SV in P T+1 The velocity of the point, V TV3 Let a be the speed of the vehicle TV3 ahead. T+1 For the current vehicle SV in P T+1 The acceleration of a point, d sv_TV3 The relative distance between the current vehicle SV and the vehicle ahead TV3, TTC th3 This is the third preset threshold.

[0100] It can be inferred that the current vehicle's SV is in P T+1 The motion state of a point is affected by the vehicle TV3 ahead, and the function f(TV3) is expressed as follows:

[0101]

[0102] In summary, based on the interactive effects of test vehicle TV1, test vehicle TV2, and preceding vehicle TV3 on the current vehicle SV during lane changing, as described in formulas (7), (9), and (11), the current vehicle SV at P T+1 The motion state of a point should be considered in balance, taking into account the effects of three functions: f(TV1), f(TV2), and f(TV3). Considering the interactive effects of test vehicle TV1, test vehicle TV2, and the vehicle ahead TV3 in the scene, the following penalty function P should be established. F(TV) :

[0103] P F(TV) =k1·f(TV1)+k2·f(TV2)+k3·f(TV3) (12)

[0104] Among them, P F(TV) Let f(TV1) be the function of the penalty term for the interaction influence of surrounding vehicles, f(TV1) be the function of the current vehicle being affected by vehicle TV1, and k1 be the influence coefficient of f(TV1); f(TV2) be the function of the current vehicle being affected by vehicle TV2, and k2 be the influence coefficient of f(TV2); and f(TV3) be the function of the current vehicle being affected by vehicle TV3, and k3 be the influence coefficient of f(TV3).

[0105] Furthermore, Huber Loss is a robust loss function that combines the advantages of Mean Squared Error (MSE) and Mean Absolute Error (MAE), and it is widely used in regression tasks (such as position prediction and numerical prediction). In this embodiment, to ensure data prediction accuracy and fast convergence during the training of the preset motion state planning model MotPreTCN, Huber Loss is considered. Based on the state information list format input to the MotPreTCN model, the Huber Loss formula is expressed as follows:

[0106]

[0107] Where M represents the actual value of the list of state information of the current vehicle at the predicted location point. This is a list of predicted current vehicle state information. δ is a threshold parameter used to control the boundary between squared loss and linear loss.

[0108] Finally, the embodiments of this application also consider penalties for passenger comfort.

[0109] Specifically, the current vehicle SV from P T Point moves to P T+1 The changes in velocity and heading angle of a point should be less than a preset threshold, which is set to M. th Then we can obtain the following constraints:

[0110] Q*(M T+1 -M T ) / ΔT≤M th (14)

[0111] in, q1 is the weight of velocity V, q2 is the weight of acceleration a, and q3 is the weight of heading angle θ; M T+1 For the current vehicle SV in P T+1 List of status information for point M T For the current vehicle SV in P T The list of point status information, ΔT is the current vehicle SV from P T Point moves to PT+1 Time taken to reach the point, M th This is a preset threshold for change.

[0112] This can also be rearranged into a penalty term function P that considers passenger comfort. P (M T+1 ):

[0113]

[0114] in, q1 is the weight of velocity V, q2 is the weight of acceleration a, and q3 is the weight of heading angle θ. v th For the threshold value of velocity V, a th For the threshold of acceleration a, θ th The threshold for the heading angle θ; ΔT is the current vehicle SV from P. T Point moves to P T+1 Time taken.

[0115] Furthermore, in this embodiment, considering formulas (12), (13), and (15), and taking into account various influencing factors in different scenarios, a loss function considering penalty terms is formulated for the preset motion state planning model MotPreTCN. P as follows:

[0116]

[0117] Among them, Loss P The default loss function is... Let P be the robust loss function, λ1 be the weighting factor for the penalty term considering passenger comfort, and P be the weighting factor for the robust loss function. F(TV) Let λ be the penalty term function for the interaction effects of surrounding vehicles, and λ2 be the weighting factor for the penalty term considering the interaction effects of surrounding vehicles. To account for the penalty factor that influences interactions with surrounding vehicles, M represents the actual value of the vehicle's state information list at the predicted location point. This provides the predicted motion state information for the current vehicle.

[0118] Therefore, this invention proposes a motion state planning method based on a temporal neural network. This method predicts and analyzes historical and current position data to accurately infer the vehicle's next position motion state. As a learning-based intelligent algorithm, this method, after being fully trained on multi-scenario datasets, possesses strong scene adaptability and can effectively cope with complex and ever-changing real-world application scenarios. In terms of data processing, this method innovatively relies solely on adjacent historical position data and current position motion state information within a short period to predict the next position motion state planning information. This feature not only greatly simplifies the network structure but also significantly reduces computational complexity, improving the algorithm's operating efficiency and real-time performance. In terms of prediction results, the next position vehicle motion state information output by this method maintains a high degree of consistency with the current and historical position motion states, effectively avoiding drastic changes in key parameters such as speed and acceleration, thereby providing drivers and passengers with a smooth and comfortable driving experience.

[0119] The motion state planning method based on a temporal convolutional neural network proposed in this application involves inputting the current vehicle's motion state information at the current moment and at least one historical moment's motion state information into a preset motion state planning model for prediction, thereby obtaining the predicted motion state information of the current vehicle. Combined with a preset loss function, lane-changing control commands for the vehicle are generated. This solves the problems of low utilization of historical data, insufficient adaptability to dynamic scenarios, and high computational complexity in the prior art, achieving accurate prediction of the vehicle's next position motion state. It effectively improves the real-time performance, accuracy, and scene adaptability of speed planning, providing users with a smooth and comfortable driving experience.

[0120] Next, with reference to the accompanying drawings, a motion state planning device based on a temporal convolutional neural network according to an embodiment of this application is described.

[0121] Figure 4 This is a block diagram of a motion state planning device based on a temporal convolutional neural network according to an embodiment of this application.

[0122] like Figure 4 As shown, the motion state planning device 10 based on a temporal convolutional neural network includes: an acquisition module 100, a prediction module 200, and a planning module 300.

[0123] The acquisition module 100 is used to acquire the motion state information of the current vehicle at the current moment and the motion state information of at least one historical moment adjacent to the current moment; the prediction module 200 is used to input the motion state information of the current moment and the motion state information of at least one historical moment into a preset motion state planning model for prediction to obtain the predicted motion state information of the current vehicle; the planning module 300 is used to combine a preset loss function and generate lane change control commands for the vehicle based on the predicted motion state information of the current vehicle.

[0124] Furthermore, in some embodiments, the preset loss function is determined by a penalty function for the impact of surrounding vehicle interactions, a robust loss function, and a user comfort penalty function.

[0125] Furthermore, in some embodiments, the function for the penalty term related to the interaction with surrounding vehicles is:

[0126] P F(TV) =k1·f(TV1)+k2·f(TV2)+k3·f(TV3);

[0127] Among them, P F(TV) Let f(TV1) be the function of the penalty term for the interaction influence of surrounding vehicles, k1 be the influence coefficient of f(TV1), f(TV1) be the function of the current vehicle being affected by vehicle TV1, k2 be the influence coefficient of f(TV2), f(TV2) be the function of the current vehicle being affected by vehicle TV2, k3 be the influence coefficient of f(TV3), and f(TV3) be the function of the current vehicle being affected by vehicle TV3.

[0128] The user comfort penalty function is:

[0129]

[0130] Among them, P P (M T+1 ) is a function for user comfort penalty terms. q1 is the weight of velocity V, q2 is the weight of acceleration a, and q3 is the weight of heading angle θ. v th For the threshold value of velocity V, a th For the threshold of acceleration a, θ th M is the threshold value for the heading angle θ. T+1 For the current vehicle at P T+1 List of status information for point M T For the current vehicle at P T The list of status information for points, where ΔT represents the current vehicle's position from point P. T Point moves to P T+1 Time taken.

[0131] Furthermore, in some embodiments, the preset loss function is:

[0132]

[0133] Among them, Loss P The default loss function is... Let P be the robust loss function, λ1 be the weighting factor for the penalty term considering passenger comfort, and P be the weighting factor for the robust loss function. F(TV) Let λ be the penalty term function for the interaction effects of surrounding vehicles, and λ2 be the weighting factor for the penalty term considering the interaction effects of surrounding vehicles. To account for the penalty factor that influences interactions with surrounding vehicles, M represents the actual value of the vehicle's state information list at the predicted location point. This provides the predicted motion state information for the current vehicle.

[0134] Furthermore, in some embodiments, the motion state information of the current vehicle at the current moment includes at least one of the current vehicle speed, the current acceleration, and the current heading angle.

[0135] It should be noted that the foregoing explanation of the motion state planning method based on temporal convolutional neural networks also applies to the motion state planning device based on temporal convolutional neural networks in this embodiment, and will not be repeated here.

[0136] The motion state planning device based on a temporal convolutional neural network proposed in this application inputs the motion state information of the current vehicle at the current moment and the motion state information at at least one historical moment into a preset motion state planning model for prediction, thereby obtaining the predicted motion state information of the current vehicle; combined with a preset loss function, a lane-changing control command for the vehicle is generated. This solves the problems of low utilization of historical data, insufficient adaptability to dynamic scenarios, and high computational complexity in the prior art, achieving accurate prediction of the vehicle's next position motion state, effectively improving the real-time performance, accuracy, and scene adaptability of speed planning, and providing users with a smooth and comfortable driving experience.

[0137] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include:

[0138] The memory 501, the processor 502, and the computer program stored on the memory 501 and capable of running on the processor 502.

[0139] When the processor 502 executes the program, it implements the motion state planning method based on the temporal convolutional neural network provided in the above embodiments.

[0140] Furthermore, electronic devices also include:

[0141] Communication interface 503 is used for communication between memory 501 and processor 502.

[0142] The memory 501 is used to store computer programs that can run on the processor 502.

[0143] The memory 501 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0144] If the memory 501, processor 502, and communication interface 503 are implemented independently, then the communication interface 503, memory 501, and processor 502 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 5 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0145] Optionally, in a specific implementation, if the memory 501, processor 502, and communication interface 503 are integrated on a single chip, then the memory 501, processor 502, and communication interface 503 can communicate with each other through an internal interface.

[0146] Processor 502 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0147] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described motion state planning method based on a temporal convolutional neural network.

[0148] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0149] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0150] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0151] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0152] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0153] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0154] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0155] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A motion state planning method based on a temporal convolutional neural network, characterized in that, Includes the following steps: Obtain the motion state information of the current vehicle at the current moment and the motion state information of at least one historical moment adjacent to the current moment; The motion state information at the current moment and the motion state information at at least one historical moment are input into a preset motion state planning model for prediction to obtain the predicted motion state information of the current vehicle. Based on a preset loss function and the predicted motion state information of the current vehicle, a lane-changing control command is generated for the vehicle.

2. The method according to claim 1, characterized in that, The preset loss function is determined by the surrounding vehicle interaction impact penalty function, the robust loss function, and the user comfort penalty function.

3. The method according to claim 2, characterized in that, The function for the penalty term related to the interaction with surrounding vehicles is: P F(TV) =k1·f(TV1)+k2·f(TV2)+k3·f(TV3); Among them, P F(TV) Let f(TV1) be the function of the penalty term for the interaction influence of surrounding vehicles, k1 be the influence coefficient of f(TV1), f(TV1) be the function of the current vehicle being affected by vehicle TV1, k2 be the influence coefficient of f(TV2), f(TV2) be the function of the current vehicle being affected by vehicle TV2, k3 be the influence coefficient of f(TV3), and f(TV3) be the function of the current vehicle being affected by vehicle TV3. The user comfort penalty function is: Among them, P P (M T+1 ) is a function for user comfort penalty terms. q1 is the weight of velocity V, q2 is the weight of acceleration a, and q3 is the weight of heading angle θ. v th For the threshold value of velocity V, a th For the threshold of acceleration a, θ th M is the threshold value for the heading angle θ. T+1 For the current vehicle at P T+1 List of status information for point M T For the current vehicle at P T The list of status information for points, where ΔT represents the current vehicle's position from point P. T Point moves to P T+1 Time taken.

4. The method according to claim 2, characterized in that, The preset loss function is: Among them, Loss P The default loss function is... Let P be the robust loss function, λ1 be the weighting factor for the penalty term considering passenger comfort, and P be the weighting factor for the robust loss function. F(TV) Let λ be the penalty term function for the interaction effects of surrounding vehicles, and λ2 be the weighting factor for the penalty term considering the interaction effects of surrounding vehicles. To account for the penalty factor that influences interactions with surrounding vehicles, M represents the actual value of the vehicle's state information list at the predicted location point. This provides the predicted motion state information for the current vehicle.

5. The method according to claim 1, characterized in that, The current vehicle's motion status information at the current moment includes at least one of the following: current vehicle speed, current acceleration, and current heading angle.

6. A motion state planning device based on a temporal convolutional neural network, characterized in that, include: The acquisition module is used to acquire the motion state information of the current vehicle at the current moment and the motion state information of at least one historical moment adjacent to the current moment; The prediction module is used to input the motion state information at the current moment and the motion state information at at least one historical moment into a preset motion state planning model for prediction, so as to obtain the predicted motion state information of the current vehicle. The planning module is used to generate lane-changing control commands for the vehicle based on the predicted motion state information of the current vehicle, combined with a preset loss function.

7. The apparatus according to claim 6, characterized in that, The preset loss function is determined by the surrounding vehicle interaction impact penalty function, the robust loss function, and the user comfort penalty function.

8. The apparatus according to claim 7, characterized in that, The function for the penalty term related to the interaction with surrounding vehicles is: P F(TV) =k1·f(TV1)+k2·f(TV2)+k3·f(TV3); Among them, P F(TV) Let f(TV1) be the function of the penalty term for the interaction influence of surrounding vehicles, k1 be the influence coefficient of f(TV1), f(TV1) be the function of the current vehicle being affected by vehicle TV1, k2 be the influence coefficient of f(TV2), f(TV2) be the function of the current vehicle being affected by vehicle TV2, k3 be the influence coefficient of f(TV3), and f(TV3) be the function of the current vehicle being affected by vehicle TV3. The user comfort penalty function is: Among them, P P (M T+1 ) is a function for user comfort penalty terms. q1 is the weight of velocity V, q2 is the weight of acceleration a, and q3 is the weight of heading angle θ. v th For the threshold value of velocity V, a th For the threshold of acceleration a, θ th M is the threshold value for the heading angle θ. T+1 For the current vehicle at P T+1 List of status information for point M T For the current vehicle at P T The list of status information for points, where ΔT represents the current vehicle's position from point P. T Point moves to P T+1 Time taken.

9. An electronic device, characterized in that, include: The memory, the processor, and the computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the motion state planning method based on a temporal convolutional neural network as described in any one of claims 1-5.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer program is executed by a processor to implement the motion state planning method based on a temporal convolutional neural network as described in any one of claims 1-5.

Images