Method for predicting response of vehicle motion state and related device

Abstract

The application relates to a method for predicting a vehicle motion state response and a related device. The method comprises the following steps: extracting a motion state coding feature of a vehicle at a current time according to a historical control amount of the vehicle before the current time and a historical motion state response amount of the vehicle in response to the historical control amount; performing feature coding on an expected control amount of the vehicle at the current time, and mapping the expected control amount into the same feature dimension as the motion state coding feature to serve as an expected control amount coding feature of the vehicle after the current time, wherein the expected control amount is used for controlling a motion state response amount of the vehicle after the current time; fusing the motion state coding feature and the expected control amount coding feature to obtain a fusion feature; and decoding the fusion feature to obtain a predicted motion state response amount of the vehicle after the current time. Through the application, the prediction of the motion state response of the vehicle is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of vehicle control, and in particular, to a method for predicting a vehicle motion state response and a related device thereof. BACKGROUND

[0002] With the development of automatic driving, a method for predicting a vehicle motion state emerges as the times require. In the field of automatic driving, the information such as the environment nearby, the vehicle and its current motion state is often obtained by a driving radar or other sensors on the vehicle to predict the future position and motion route of the nearby vehicle, so as to control or assist the driving of the vehicle.

[0003] The inventors have found in the research process that the relationship between the motion state response of the vehicle and the control quantity presents complexity rather than always maintaining a linear relationship. For example, when the vehicle is at different speeds and angular velocities, the same control quantity applied to the vehicle may result in different state responses of the vehicle speed or angular velocity. Accurate prediction of the motion state response of the vehicle helps the driver or the automatic driving system to accurately apply a control quantity according to the current motion state of the vehicle, so that the vehicle can run in the motion state expected by the driver or the automatic driving system after the control quantity is applied, thereby improving the driving safety.

[0004] However, there is no effective solution to the problem of how to predict the motion state response of the vehicle. SUMMARY

[0005] The method for predicting a vehicle motion state response and the related device thereof provided by the embodiments at least solve the problem of how to predict the motion state response of the vehicle.

[0006] A method for predicting a vehicle motion state response, comprising:

[0007] extracting a motion state encoding feature of the vehicle at a current time according to a historical control quantity of the vehicle before the current time and a historical motion state response quantity of the vehicle in response to the historical control quantity;

[0008] encoding an expected control quantity of the vehicle at the current time and mapping it to the same feature dimension as the motion state encoding feature as an expected control quantity encoding feature of the vehicle after the current time, wherein the expected control quantity is used to control a motion state response quantity of the vehicle after the current time;

[0009] fusing the motion state encoding feature and the expected control quantity encoding feature to obtain a fusion feature;

[0010] decoding the fusion feature to obtain a predicted motion state response quantity of the vehicle after the current time.

[0011] In some embodiments among them,

[0012] Before extracting the motion state encoding feature of the vehicle at the current time according to the historical control amount of the vehicle before the current time and the historical motion state response amount generated by the vehicle in response to the historical control amount, the method further comprises: sampling the historical control amount and the historical motion state response amount to obtain a historical control amount sequence and a historical motion state response amount sequence, and fusing the historical control amount sequence and the historical motion state response amount sequence according to the sampling time corresponding to each element in the historical control amount sequence and the historical motion state response amount sequence to obtain a fused sequence;

[0013] Extracting the motion state encoding feature of the vehicle at the current time according to the historical control amount of the vehicle before the current time and the historical motion state response amount generated by the vehicle in response to the historical control amount comprises: extracting the motion state encoding feature of the vehicle at the current time according to the fused sequence.

[0014] In some embodiments among them, the motion state encoding feature is extracted by a first encoder based on an artificial neural network, the first encoder comprising an input layer, a hidden layer and an output layer; wherein extracting the motion state encoding feature of the vehicle at the current time according to the fused sequence comprises:

[0015] Inputting the fused sequence from the input layer;

[0016] For each hidden layer:

[0017] When the layer above the current hidden layer is the input layer, obtaining the hidden state of the node corresponding to the t-th time in the current hidden layer according to the input data of the n+1 nodes corresponding to the t-th time to the t-n-th time in the input layer and the output result of the node corresponding to the t-1-th time in the output layer;

[0018] When the layer above the current hidden layer is the hidden layer, obtaining the hidden state of the node corresponding to the t-th time in the current hidden layer according to the hidden state of the n+1 nodes corresponding to the t-th time to the t-n-th time in the hidden layer of the layer above;

[0019] When the layer below the current hidden layer is the output layer, processing the hidden state of the current hidden layer by the node corresponding to the t-th time in the output layer and obtaining the output result as the motion state encoding feature of the vehicle at the t-th time.

[0020] In some embodiments, the motion state encoding feature is extracted by a first artificial neural network-based encoder, which includes an input layer, hidden layers, and an output layer; wherein the motion state encoding feature of the vehicle at the current time is extracted according to the fusion sequence, including:

[0021] inputting the fusion sequence from the input layer;

[0022] for each hidden layer:

[0023] when the previous layer of the current hidden layer is the input layer, obtaining the hidden state of the node corresponding to the t-th time in the current hidden layer according to the input data of the n+1 nodes corresponding to the t-th time to the t-n-th time in the input layer and the output result of the node corresponding to the t-1-th time in the output layer;

[0024] when the previous layer of the current hidden layer is a hidden layer, selecting n+1 hidden states in the hidden states of the nodes corresponding to the t-th time to the t-n-th time in the previous hidden layer according to equal time intervals, and obtaining the hidden state of the node corresponding to the t-th time in the current hidden layer according to the selected n+1 hidden states, wherein the time interval increases layer by layer with the number of hidden layers;

[0025] when the next layer of the current hidden layer is the output layer, processing the hidden state of the current hidden layer by the node corresponding to the t-th time in the output layer to obtain an output result as the motion state encoding feature of the vehicle at the t-th time.

[0026] In some embodiments, after processing the hidden state of the current hidden layer by the node corresponding to the t-th time in the output layer to obtain an output result as the motion state encoding feature of the vehicle at the t-th time, the method further includes:

[0027] when the output layer includes nodes corresponding to future times, pruning the output results of the nodes corresponding to the future times from the output results of the output layer; and

[0028] cutting a preset number of output results between the current time and the historical time from the output results of the output layer as the motion state encoding feature.

[0029] In some embodiments, the expected control quantity encoding feature is extracted by a second artificial neural network-based encoder, which includes an input layer, hidden layers, and an output layer; wherein the expected control quantity is feature-encoded and mapped to the same feature dimension as the motion state encoding feature as the expected control quantity encoding feature of the vehicle after the current time, including:

[0030] sampling the expected control amount to obtain an expected control amount sequence;

[0031] inputting the expected control amount sequence from the input layer;

[0032] for each hidden layer:

[0033] when the previous layer of the current hidden layer is the input layer, obtaining the hidden state of each node in the current hidden layer according to the input data of all nodes in the input layer;

[0034] when the previous layer of the current hidden layer is a hidden layer, obtaining the hidden state of each node in the current hidden layer according to the hidden state of all nodes in the hidden layer of the previous layer;

[0035] when the next layer of the current hidden layer is the output layer, processing the hidden state of each node in the current hidden layer by the output layer and obtaining an output result as the expected control amount encoding feature, wherein the number of nodes of the output layer is the same as the feature dimension number of the motion state encoding feature.

[0036] In some embodiments thereof, fusing the motion state encoding feature and the expected control amount encoding feature to obtain a fusion feature includes:

[0037] fusing the motion state encoding feature and the expected control amount encoding feature through a parallel strategy to obtain the fusion feature.

[0038] In some embodiments thereof, the predicted motion state response amount is decoded by a decoder based on an artificial neural network; the decoder includes an input layer, a hidden layer and an output layer; wherein decoding the fusion feature to obtain the predicted motion state response amount of the vehicle after the current time includes:

[0039] inputting the fusion feature from the input layer;

[0040] obtaining the hidden state of the node corresponding to the t-th time in the hidden layer according to the input data of the node corresponding to the t-th time in the input layer, the output result of the node corresponding to the t-1-th time in the output layer and the hidden state of the node corresponding to the t-1-th time in the hidden layer;

[0041] obtaining the output result of the node corresponding to the t-th time in the output layer according to the hidden state of the node corresponding to the t-th time in the hidden layer and the output result of the node corresponding to the t-1-th time in the output layer, as the predicted motion state response amount of the vehicle at the t-th time.

[0042] A prediction device for a vehicle motion state response includes:

[0043] a first encoder configured to extract a motion state encoding feature of the vehicle at the current time according to historical control amounts of the vehicle before the current time and historical motion state response amounts of the vehicle in response to the historical control amounts;

[0044] a second encoder configured to encode an expected control amount of the vehicle at the current time and map the expected control amount into a same feature dimension as the motion state encoding feature as an expected control amount encoding feature of the vehicle after the current time, wherein the expected control amount is used to control a motion state response amount of the vehicle after the current time;

[0045] a fusion module configured to fuse the motion state encoding feature and the expected control amount encoding feature to obtain a fusion feature;

[0046] a decoder configured to decode the fusion feature to obtain a predicted motion state response amount of the vehicle after the current time.

[0047] An electronic device includes a processor and a memory storing a program including instructions that, when executed by the processor, cause the processor to perform the above method.

[0048] A non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the above method.

[0049] The vehicle motion state response prediction method and related device provided by the embodiment extract a motion state encoding feature of the vehicle at the current time according to historical control amounts of the vehicle before the current time and historical motion state response amounts of the vehicle in response to the historical control amounts, encode an expected control amount of the vehicle at the current time and map the expected control amount into a same feature dimension as the motion state encoding feature as an expected control amount encoding feature of the vehicle after the current time, wherein the expected control amount is used to control a motion state response amount of the vehicle after the current time, fuse the motion state encoding feature and the expected control amount encoding feature to obtain a fusion feature, and decode the fusion feature to obtain a predicted motion state response amount of the vehicle after the current time, thereby achieving prediction of the motion state response of the vehicle.

[0050] The details of one or more embodiments of the application are set forth in the accompanying drawings and the description below. Other features, objects, and advantages of the application will be apparent from the description of certain embodiments of the application and from the claims. BRIEF DESCRIPTION OF DRAWINGS

[0051] The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the application. In the drawings:

[0052] Figure 1 is a flowchart of a method for predicting a vehicle motion state response according to an embodiment of the present disclosure.

[0053] Figure 2 is a schematic diagram of a method for predicting a vehicle motion state response according to an embodiment of the present disclosure.

[0054] Figure 3 is a structural schematic diagram of a first encoder according to an embodiment of the present disclosure.

[0055] Figure 4 is a structural schematic diagram of a causal convolutional artificial neural network according to an embodiment of the present disclosure.

[0056] Figure 5 is a schematic diagram of boundary padding and cropping of a causal convolutional artificial neural network according to an embodiment of the present disclosure.

[0057] Figure 6 is a structural schematic diagram of a second encoder according to an embodiment of the present disclosure.

[0058] Figure 7 is a structural schematic diagram of a method for predicting a vehicle motion state response according to an embodiment of the present disclosure.

[0059] Figure 8 is a schematic diagram of input information and output information of a method for predicting a vehicle motion state response according to an embodiment of the present disclosure.

[0060] Figure 9 is a structural schematic diagram of a method for predicting a vehicle motion state response according to an embodiment of the present disclosure.

[0061] Figure 10 is a schematic diagram of a workflow of a conditional decoder according to an embodiment of the present disclosure.

[0062] Figure 11 is a structural schematic diagram of an electronic device according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0063] Embodiments of the present disclosure will be described in more detail with reference to the accompanying drawings. While the present disclosure is susceptible to various modifications and alternative forms, specific embodiments thereof are shown by way of example in the drawings and will herein be described in detail. It should be understood, however, that there is no intent to limit the present disclosure to the particular embodiments disclosed, but on the contrary, the disclosure is to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the present disclosure. It will be understood that the drawings and the following detailed description are not intended to limit the scope of the present disclosure to the particular embodiments described in the following detailed description.

[0064] It should be understood that the various steps of the method implementation of the present embodiment can be performed in different orders and / or in parallel. In addition, the method implementation can include additional steps and / or omit the steps shown. The scope of the present embodiment is not limited in this respect.

[0065] The term "comprising" and variations thereof as used herein are open-ended, that is "including but not limited to". The term "based on" is "based, at least in part, on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Related terms are defined in the following description.

[0066] It should be noted that the modification of "one" or "a plurality of" mentioned in the present embodiment is illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise explicitly indicated in the context, it should be understood as "one or more".

[0067] The present embodiment provides a prediction method for vehicle motion state response, and a device for executing the prediction method can be arranged in a cloud service cluster or a domain controller (DCU) of an advanced driving assistance system (ADAS). The prediction method provided by the present embodiment can be implemented by a prediction device based on artificial neural network. The prediction device can be used for online implementation of prediction, and used as a dynamic model in a model predictive control algorithm; the prediction device can also be used for offline simulation, and used as a vehicle dynamics feedback model in a simulator.

[0068] Vehicle control is usually in the form of feedback control. For example, in modern control theory of vehicles, the state space model of the internal characteristics of the vehicle system is used to describe the object, and the measurable output signal and the internal state of the system are used as feedback signals to adjust the motion state of the vehicle, so that the motion state response of the vehicle reaches the desired state. The vehicle control in the related art is usually represented by the state space description of a linear constant continuous system to represent the relationship between the control amount of the vehicle and the motion state response thereof. The state space description of the linear constant continuous system includes two parts: a state equation and an output equation. The state equation describes the relationship between the next motion state of the system and the current motion state and the current control amount. However, how to set the state equation to achieve the optimal control of the vehicle control is the core problem of the optimization of the linear constant continuous system.

[0069] Unlike the related art, in the present embodiment, the state equation is no longer used to describe the relationship between the next motion state of the system and the current motion state and the current control amount, but a feature encoding and decoding based on an encoder-decoder architecture artificial neural network model is used to achieve the above purpose. Through the prediction of the vehicle motion state response by the artificial neural network, the feedforward control of the vehicle is realized.

[0070] Figure 1 is a flowchart of a prediction method of a vehicle motion state response of the embodiment, as shown in the flowchart, the flow includes the following steps: Figure 1

[0071] Step S101, according to the historical control amount of the vehicle before the current time, and the historical motion state response amount generated by the vehicle in response to the historical control amount, the motion state encoding features of the vehicle at the current time are extracted.

[0072] Step S102, the expected control amount of the vehicle at the current time is encoded and mapped into the same feature dimension as the motion state encoding features, as the expected control amount encoding features of the vehicle after the current time, wherein the expected control amount is used to control the motion state response amount of the vehicle after the current time.

[0073] Step S103, the motion state encoding features and the expected control amount encoding features are fused to obtain the fusion features.

[0074] Step S104, the fusion features are decoded to obtain the predicted motion state response amount of the vehicle after the current time.

[0075] In the above step S101, the control amount of the vehicle and the motion state response amount of the vehicle can be obtained by a sampling device according to a certain sampling interval from each motion and state sensor of the vehicle, wherein the control amount of the vehicle includes but is not limited to each control amount for controlling the steering, speed and acceleration of the vehicle, such as steering wheel rotation angle, throttle size, etc. The motion state response amount of the vehicle includes but is not limited to the linear speed, acceleration and angular velocity of the vehicle, etc. In addition, the motion state response amount of the vehicle can also include tire pressure and other feature data associated with the motion state of the vehicle.

[0076] The above steps S101, S102 and S104 can be realized by using a prediction device composed of a time series processing artificial neural network.

[0077] The control amount and motion state response amount of the vehicle are usually continuous values. Before the above step S101, the historical control amount and the historical motion state response amount are sampled to obtain the historical control amount sequence and the historical motion state response amount sequence, and the historical control amount sequence and the historical motion state response amount sequence are fused according to the sampling time corresponding to each element in the historical control amount sequence and the historical motion state response amount sequence to obtain the fusion sequence. Then in step S101, the artificial neural network can take the fusion sequence as the input data to extract the motion state encoding features of the vehicle at the current time.

[0078] ​Similarly, in step S102, the desired control quantity sequence that can be processed by the prediction device described above is also obtained by sampling the desired control quantity.

[0079] In the sampling of the historical control quantity and the historical motion state response quantity, the sampling rate of the control quantity and the motion state response quantity is consistent or substantially consistent, for example, the same number of sampling results are uniformly collected within the same time range. After sampling, the control quantity sequence and the motion state response quantity sequence are obtained, and each element in the sequence is identified by a time stamp or sorted in chronological order.

[0080] In the present embodiment, the current time refers to the reference time for predicting the vehicle motion state response, and the historical control quantity sequence and the historical motion state response quantity sequence both refer to a certain number of sampling results before the reference time. Each certain number of continuous sampling results is referred to as a time step in the present embodiment, and the length of each time step can be set as needed.

[0081] Figure 2 is a schematic diagram of the prediction method of the vehicle motion state response based on the artificial neural network prediction device of the present embodiment, as shown in Figure 2 The prediction device adopts an encoder-decoder architecture, wherein the prediction device includes two encoders and one decoder, and the encoders and the decoder are artificial neural networks that can be used to process time series, such as but not limited to: multilayer perceptron, recurrent artificial neural network (RNN), long short-term memory network (LSTM), or a variant of the above artificial neural network.

[0082] Referring back to Figure 2 The first encoder in the prediction device is used to extract and encode the fixed-length or variable-length fusion sequence to obtain a fixed-length motion state coding feature, wherein the first encoder includes a plurality of encoding units connected in sequence, and the fusion sequence is input into the encoder in chronological order. Each encoding unit determines the output data of the current encoding unit according to the input data of the current encoding unit and the output data of the previous encoding unit. Therefore, the output data of the last encoding unit has encoded the information of the input data of all the encoding units, i.e., the motion state coding feature of the above embodiment.

[0083] Similar to the first encoder, the second encoder is used to extract and encode the desired control quantity sequence of fixed length or variable length to obtain the desired control quantity encoding feature of fixed length. The second encoder also includes a plurality of encoding units connected in sequence, and the desired control quantity sequence is input into the encoder in time sequence, and each encoding unit determines the output data of the current encoding unit according to the input data of the current encoding unit and the output data of the previous encoding unit. Therefore, the output data of the last encoding unit has encoded the information of the input data of all the encoding units, that is, the desired control quantity encoding feature of the above-mentioned embodiment.

[0084] It should be noted that the number of encoding units of the two encoders described above can be different, but the feature dimensions of the encoding features output by the two encoders are the same.

[0085] In this embodiment, the motion state encoding feature and the desired control quantity encoding feature are not directly input into the decoder, but are fused into a fusion feature by a fusion module, which is used as the input of the decoder. The decoder is used to decode (i.e., decode) the fixed-length or variable-length fusion feature into a fixed-length predicted motion state response quantity sequence. The decoder also includes a plurality of decoding units connected in sequence, and the fusion feature will be used as the input data of each decoding unit. At the same time, each decoding unit calculates the hidden state and output result of the current decoding unit according to the output result of the previous decoding unit and the hidden state, and the fusion feature input into the current decoding unit. The output result of each decoding unit corresponds to a predicted motion state response quantity at a future time.

[0086] The motion state response of the vehicle at each time has a causal relationship with the vehicle motion state response in a period of time before that and the control quantity in a period of time before that. In order to extract this causal relationship, in this embodiment, the motion state encoding feature is extracted by the first encoder based on artificial neural network. Figure 3 is a structural schematic diagram of the first encoder of this embodiment, as Figure 3 shown, the first encoder includes an input layer, a hidden layer and an output layer. Among them, the number of hidden layers can be one or more.

[0087] It should be noted that in Figure 3 in order to clearly show the connection between nodes, only the connection relationship of the nodes related to the output result Ht of the tth time is shown, and the connection relationship between other nodes is omitted.

[0088] Referring to Figure 3 , the above step S103 includes: inputting the fusion sequence from the input layer; for each hidden layer:

[0089] ① When the layer above the current hidden layer is the input layer, the hidden state of the node corresponding to the t-th moment in the current hidden layer is obtained according to the input data of n+1 nodes corresponding to the t-th moment to the t-n-th moment in the input layer and the output result of the node corresponding to the t-1-th moment in the output layer;

[0090] ② When the layer above the current hidden layer is the hidden layer, the hidden state of the node corresponding to the t-th moment in the current hidden layer is obtained according to the hidden states of n+1 nodes corresponding to the t-th moment to the t-n-th moment in the hidden layer above the current hidden layer;

[0091] ③ When the layer below the current hidden layer is the output layer, the output result of the node corresponding to the t-th moment in the output layer is obtained by processing the hidden state of the current hidden layer, and is taken as the motion state encoding feature of the vehicle at the t-th moment.

[0092] Generally, in the output layer, the hidden state output by the node corresponding to each moment before the t-th moment in the last hidden layer can be transformed into the output result of the node corresponding to the t-th moment in the output layer by a self-defined function. However, since the hidden state of the node corresponding to the t-th moment in the last hidden layer already contains the hidden states of all nodes before the t-th moment, in the embodiment, the hidden state of the node corresponding to the t-th moment in the last hidden layer can be directly taken as the output result of the node corresponding to the t-th moment in the output layer.

[0093] In the related art, the features of time series data are usually extracted by using a recurrent artificial neural network (RNN) or a long short-term memory network (LSTM). However, there are a large number of full connections in the RNN and the LSTM, that is, in addition to the forward connection from the node representing the current time to the node representing the future time, there are a large number of backward connections from the node representing the future time to the node representing the current time in each layer. The aforementioned causal relationship refers to the fact that the output result of the current moment only depends on the input data or the output result of the current moment or the historical moment, and is irrelevant to the input of the future moment. Therefore, the causal relationship cannot include the influence of the future state on the current state. Due to the existence of the backward connection, if the causal features are extracted by using the RNN or the LSTM, the network parameters will be too large to converge, or the training process will consume a large amount of computing resources and time cost.

[0094] Therefore, in this embodiment, the first encoder is a causal convolutional artificial neural network. Compared to RNNs or LSTMs, causal convolutional artificial neural networks remove or mask all backward connections, retaining only forward connections. This ensures that input data received by nodes in future timeframes no longer affects the output at the current timeframe, thus maintaining the network's strict causal properties. Furthermore, by removing or masking backward connections, causal convolutional artificial neural networks are more lightweight than RNNs or LSTMs, with significantly fewer network parameters compared to RNNs or LSTMs with the same number of nodes per layer and number of layers, making them easier to train.

[0095] Furthermore, compared to commonly used recurrent artificial neural networks (RNNs), causal convolutional artificial neural networks can process all historical nodes in parallel at once, without requiring sequential iteration. After multi-layer causal convolution calculations, the field of view of the causal convolutional artificial neural network can completely cover a certain time length, and the output encoded state feature vector is the motion state encoding feature mentioned above.

[0096] Furthermore, considering the normal response delay of vehicle actuators, the method in this embodiment for predicting vehicle operating state response needs to have the ability to extract features from sufficiently long historical information; that is, the sequence length of the aforementioned historical control quantity sequence and historical operating state response quantity sequence may be as long as tens or even hundreds of times. Only using... Figure 3 The causal convolutional artificial neural network shown has a receptive field of view of 7 input nodes for each output node, meaning the sequence length is 7. If it is necessary to increase the receptive field... Figure 3 The development of causal convolutional artificial neural networks requires increasing the number of hidden layers. Increasing the number of hidden layers involves enlarging the network parameters and structure, which not only leads to training difficulties but may also cause gradient vanishing due to excessive network depth, preventing the network from converging. Therefore, in this embodiment, the first encoder can be a dilated causal convolutional artificial neural network, also known as a dilated causal convolutional artificial neural network. The difference between this dilated causal convolutional artificial neural network and the original causal convolutional artificial neural network lies in the different dilation rates used between its neural network layers. Figure 4 This is a schematic diagram of the structure of the dilated causal convolutional artificial neural network in this embodiment, as shown below. Figure 4 As shown, this artificial neural network is... Figure 3 The causal convolutional artificial neural network shown has the same number of layers, convolutional kernels, and nodes in each layer. The difference lies in the addition of holes in different layers. Depending on the size of the holes, the dilation rate increases progressively from top to bottom, at 1, 2, and 4 respectively. This increases the receptive field of the artificial neural network in the time dimension to 15 input nodes, meaning each output node can sense 15 pieces of historical information, including the current time and previous times. Compared to...Figure 3 The causal convolution artificial neural network shown in the figure, the hollow causal convolution artificial neural network of the embodiment achieves the purpose of increasing the receptive field without loss of information, without increasing the amount of parameters and computation.

[0097] Reference Figure 4 The step S103 includes: inputting the fusion sequence from the input layer; for each hidden layer:

[0098] ① When the previous layer of the current hidden layer is the input layer, the hidden state of the node corresponding to the t-th moment in the current hidden layer is obtained according to the input data of the n+1 nodes corresponding to the t-th moment to the t-n-th moment in the input layer and the output result of the node corresponding to the t-1-th moment in the output layer;

[0099] ② When the previous layer of the current hidden layer is the hidden layer, n+1 hidden states are selected in the hidden states of the nodes corresponding to the t-th moment to the t-n-th moment in the previous layer according to equal time intervals, and the hidden state of the node corresponding to the t-th moment in the current hidden layer is obtained according to the selected n+1 hidden states, wherein the time interval increases layer by layer with the number of hidden layers;

[0100] ③ When the next layer of the current hidden layer is the output layer, the hidden state of the current hidden layer is processed by the node corresponding to the t-th moment in the output layer to obtain an output result, which is taken as the motion state encoding feature of the vehicle at the t-th moment.

[0101] Generally, in the output layer, the hidden state output by the node corresponding to each moment within the field of view before the t-th moment in the last hidden layer can be transformed into the output result of the node corresponding to the t-th moment in the output layer by a self-defined function. However, since the hidden state of the node corresponding to the t-th moment in the last hidden layer already contains the hidden states of all nodes before the t-th moment, in the embodiment, the hidden state of the node corresponding to the t-th moment in the last hidden layer can be directly taken as the output result of the node corresponding to the t-th moment in the output layer.

[0102] In the prediction method of the motion state response of the vehicle in the embodiment, the motion state response to be predicted only depends on the historical motion state response, the historical control amount and the expected control amount at the current moment, and the motion state response of the vehicle at the future moment is unknown at the current moment of prediction, so the causal relationship between data needs to be considered when constructing the model. In the embodiment, the artificial neural network is also padded. Figure 5 is a schematic diagram of the padding and cutting of the causal convolution artificial neural network in the embodiment. It should be noted that in Figure 5 , the hidden layer is omitted for simplicity. Continue to refer to Figure 5 ​In related technologies, boundary filling methods are usually symmetrical, therefore, in Figure 5 After padding two nodes on the left side of the neural network input layer, two nodes are also padded on the right side simultaneously. These two nodes on the right correspond to nodes at future time points. The causal convolution outputs of these two future time point nodes are causal convolution results for future time points, not for the current or historical time points. Therefore, they cannot be used to characterize the motion state encoding features at the current time point. To address this, in this embodiment, after using a causal convolutional artificial neural network with symmetrically padded input layers to perform causal convolution processing on the historical control quantity sequence and the historical motion state response sequence to obtain the causal convolution processing result, this result is not directly used as the motion state encoding feature for the current time point. Instead, if the output layer includes nodes corresponding to future time points, the output results of the future time point nodes are removed from the output results of the output layer to ensure that the motion state encoding features are causally related only to the historical control quantity and the historical motion state response.

[0103] Furthermore, in the predictive application of vehicle motion state response, the further back in time the historical control variables and historical motion state response variables are from the current moment, the smaller their impact on future motion state response variables. Therefore, a predetermined number of output results between the current moment and historical moments can be extracted from the output layer of the artificial neural network as motion state encoding features. The aforementioned predetermined number can be any number from 5 to 10.

[0104] In step S102 above, the desired control quantity encoding features can also be extracted using the same artificial neural network as the first encoder. However, in the field of vehicle control, the feature dimension of the control quantity is lower than that of the motion state response quantity. Therefore, the desired control quantity encoding features can be extracted and encoded using RNN, LSTM, or multilayer perceptron, while maintaining the same feature dimension for both after encoding.

[0105] Taking a multilayer perceptron as an example, Figure 6 This is a schematic diagram of the structure of the second encoder in this embodiment, as shown below. Figure 6 As shown, the second encoder includes an input layer, a hidden layer, and an output layer. The number of hidden layers can be one or more.

[0106] Based on the above-described multilayer perceptron, step S102 includes: sampling the desired control quantity to obtain a desired control quantity sequence; inputting the desired control quantity sequence from the input layer; for each hidden layer:

[0107] ① When the layer above the current hidden layer is the input layer, obtain the hidden state of each node in the current hidden layer based on the input data of all nodes in the input layer;

[0108] ii. when the previous layer of the current hidden layer is a hidden layer, obtaining the hidden state of each node in the current hidden layer according to the hidden states of all nodes in the hidden layer of the previous layer;

[0109] iii. when the next layer of the current hidden layer is an output layer, processing the hidden states of each node in the current hidden layer by the output layer and obtaining an output result as the expected control quantity encoding feature, wherein the number of nodes of the output layer is the same as the feature dimension number of the motion state encoding feature.

[0110] In step S103, the motion state encoding feature and the expected control quantity encoding feature are fused by a feature-level fusion manner. The feature-level fusion manner includes but is not limited to a concat manner and an add manner. The concat manner directly connects the two features, and the feature dimension of the connected feature is the sum of the two features. The add manner adopts a parallel strategy to combine the two features into a feature, and the feature dimension of the composite feature is the same as that of the two features before fusion. Since the feature dimensions of the motion state encoding feature and the expected control quantity encoding feature are the same in this embodiment, the motion state encoding feature and the expected control quantity encoding feature are fused by the parallel strategy in step S103 to obtain a fused feature.

[0111] In this embodiment, the predicted motion state response quantity is decoded by a decoder based on an artificial neural network. The decoder can be an RNN, an LSTM or a causal convolution artificial neural network. The decoder includes an input layer, a hidden layer and an output layer.

[0112] With reference to the decoder shown in Figure 2 , step S104 includes: inputting the fused feature from the input layer; obtaining the hidden state of the node corresponding to the t-th moment in the hidden layer according to the input data of the node corresponding to the t-th moment in the input layer, the output result of the node corresponding to the t-1-th moment in the output layer and the hidden state of the node corresponding to the t-1-th moment in the hidden layer; and obtaining the output result of the node corresponding to the t-th moment in the output layer as the predicted motion state response quantity of the vehicle at the t-th moment according to the hidden state of the node corresponding to the t-th moment in the hidden layer and the output result of the node corresponding to the t-1-th moment in the output layer.

[0113] The embodiment also provides a vehicle motion state response prediction device, Figure 7 is a structural schematic diagram of the vehicle motion state response prediction device of the embodiment, as shown in Figure 7 , the vehicle motion state response prediction device includes:

[0114] The first encoder 71 is configured to extract the motion state encoding feature of the vehicle at the current time point according to the historical control amount of the vehicle before the current time point and the historical motion state response amount of the vehicle in response to the historical control amount.

[0115] The second encoder 72 is configured to encode the expected control amount of the vehicle at the current time point and map the expected control amount into the same feature dimension as the motion state encoding feature as the expected control amount encoding feature of the vehicle after the current time point, wherein the expected control amount is used to control the motion state response amount of the vehicle after the current time point.

[0116] The fusion module 73 is configured to fuse the motion state encoding feature and the expected control amount encoding feature to obtain the fusion feature.

[0117] The decoder 74 is configured to decode the fusion feature to obtain the predicted motion state response amount of the vehicle after the current time point.

[0118] In some embodiments, the prediction device further comprises a preprocessing module configured to sample the historical control amount and the historical motion state response amount to obtain a historical control amount sequence and a historical motion state response amount sequence, and fuse the historical control amount sequence and the historical motion state response amount sequence according to the sampling time corresponding to each element in the historical control amount sequence and the historical motion state response amount sequence to obtain a fusion sequence.

[0119] In some embodiments, the first encoder 71 comprises an input layer, a hidden layer and an output layer. The input layer of the first encoder 71 is configured to obtain the fusion sequence. For each hidden layer: when the previous layer of the current hidden layer is the input layer, the hidden state of the node corresponding to the t-th time point in the current hidden layer is obtained according to the input data of the n+1 nodes corresponding to the t-th time point to the t-n-th time point in the input layer and the output result of the node corresponding to the t-1-th time point in the output layer; when the previous layer of the current hidden layer is the hidden layer, the hidden state of the node corresponding to the t-th time point in the current hidden layer is obtained according to the hidden state of the n+1 nodes corresponding to the t-th time point to the t-n-th time point in the hidden layer of the previous layer; and when the next layer of the current hidden layer is the output layer, the output result is obtained by processing the hidden state of the current hidden layer by the node corresponding to the t-th time point in the output layer, and the output result is taken as the motion state encoding feature of the vehicle at the t-th time point.

[0120] In some embodiments, the input layer of the first encoder 71 is configured to obtain the fusion sequence. For each hidden layer: when the previous layer of the current hidden layer is the input layer, the hidden state of the node corresponding to the t-th time in the current hidden layer is obtained according to the input data of the n+1 nodes corresponding to the t-th time to the t-n-th time in the input layer and the output result of the node corresponding to the t-1-th time in the output layer; when the previous layer of the current hidden layer is a hidden layer, n+1 hidden states are selected from the hidden states of the nodes corresponding to the t-th time to the t-n-th time in the hidden layer of the previous layer according to equal time intervals, and the hidden state of the node corresponding to the t-th time in the current hidden layer is obtained according to the n+1 selected hidden states, wherein the time interval increases layer by layer with the number of hidden layers; when the next layer of the current hidden layer is the output layer, the hidden state of the current hidden layer is processed by the node corresponding to the t-th time in the output layer to obtain an output result, which is taken as the motion state encoding feature of the vehicle at the t-th time.

[0121] In some embodiments, each layer of the first encoder includes a plurality of nodes corresponding to future time. The prediction device of the vehicle motion state response further includes a pruning module, wherein the pruning module is configured to: in the case that the output layer includes nodes corresponding to future time, pruning the output result of the node corresponding to the future time from the output result of the output layer; and from the output result of the output layer, cutting out a preset number of output results between the current time and the historical time as the motion state encoding feature.

[0122] In some embodiments, the second encoder 72 includes an input layer, hidden layers and an output layer. The input layer of the second encoder 72 is configured to obtain the expected control quantity sequence. For each hidden layer: when the previous layer of the current hidden layer is the input layer, the hidden state of each node in the current hidden layer is obtained according to the input data of all nodes in the input layer; when the previous layer of the current hidden layer is a hidden layer, the hidden state of each node in the current hidden layer is obtained according to the hidden state of all nodes in the hidden layer of the previous layer; when the next layer of the current hidden layer is the output layer, the hidden state of each node in the current hidden layer is processed by the output layer to obtain an output result, which is taken as the expected control quantity encoding feature, wherein the number of nodes of the output layer is the same as the feature dimension number of the motion state encoding feature.

[0123] In some embodiments, the fusion module 73 is configured to fuse the motion state encoding feature and the expected control quantity encoding feature through a parallel strategy to obtain the fusion feature.

[0124] In some embodiments, the decoder 74 includes an input layer, a hidden layer, and an output layer. The input layer of the decoder is used to acquire fused features. The decoder 74 is also used to obtain the hidden state of the node corresponding to time t in the hidden layer based on the input data of the node corresponding to time t in the input layer, the output result of the node corresponding to time t-1 in the output layer, and the hidden state of the node corresponding to time t-1 in the hidden layer; and to obtain the output result of the node corresponding to time t in the output layer based on the hidden state of the node corresponding to time t in the hidden layer and the output result of the node corresponding to time t-1 in the output layer, as the predicted motion state response of the vehicle at time t.

[0125] In some of these embodiments, the first encoder 71 is a causal convolutional artificial neural network, the second encoder 72 is a multilayer perceptron, and the decoder is a recurrent artificial neural network.

[0126] In some of these embodiments, the first encoder is a dilated causal convolutional artificial neural network, wherein the dilation rate of each layer of the dilated causal convolutional artificial neural network increases progressively.

[0127] Figure 8 This is a schematic diagram of the input and output information of the vehicle motion state response prediction device in this embodiment, as shown below. Figure 8 As shown, the vehicle motion state response prediction device described above has two inputs: one is M frames of historical information, which includes historical control quantity sequences and historical motion state response quantity sequences, used to characterize the current state characteristics of the vehicle. Each frame represents a sampled value at a given moment. The other input is N frames of expected control quantity sequences, used to characterize the future state change conditions of the vehicle. The output of the vehicle motion state response prediction device is the predicted vehicle motion state response sequence for the next N frames.

[0128] Furthermore, if the aforementioned prediction device is divided according to its functional modules, Figure 7 It can also be represented as Figure 9 The specific structure is shown. For example... Figure 9 As shown, the prediction device can be divided into two parts: a causal encoder (CE) and a conditional decoder (CD). The causal encoder is the first encoder 71 mentioned above. Figure 9 The conditional decoder, labeled as causal conv subnet, consists of the second encoder 72, fusion module 73, and decoder 74 mentioned above. Figure 9 In this diagram, the second encoder 72 is a multilayer perceptron (MLP), and the decoder 74 is represented as a prediction head.

[0129] Figure 10is a schematic diagram of the workflow of the conditional decoder of the embodiment, as Figure 10 As shown, the CE is responsible for extracting the inherent correlation and characteristics in the historical time series data, and outputs the encoding information representing the current state of the vehicle to the CD. The CD is responsible for extracting the conditional characteristics (based on future control quantities) and fusing with the current state of the vehicle, and finally predicting the future vehicle response. The above prediction device can predict one time step unit each time it is calculated forwardly, and after the predicted value is used to update the historical value, multiple time steps of prediction can be obtained through iteration.

[0130] In the above embodiment, the first encoder, the second encoder and the decoder are obtained through end-to-end joint training. For example, the above prediction device processes the actual driving data of the vehicle into a sample set, constructs an artificial neural network model composed of a causal encoder (CE) and a conditional decoder (CD), inputs the sample set into the CE-CD model for training until the loss function converges to end the training. The prediction of the future motion state of the vehicle through the CE-CD model can achieve high prediction accuracy and provide a more accurate reference for the vehicle dynamics response.

[0131] Considering that there should be no abnormal value in the response prediction data, in the model training, the cost function adopts L2 norm (MSE) to minimize the error between the predicted response value and the actual response value. If the prediction length is selected to be large, the loss function can be adjusted to a non-equal weight form, that is, the closer the time interval of the prediction point to the current time, the greater the error weight of the point, and vice versa, which is beneficial to practical application.

[0132] The training process uses the Adam optimization algorithm to update the network weights, and the learning rate is decayed (75%) at equal intervals according to the iteration rounds. The total parameter amount of the model is 151kb, which has the characteristics of lightweight and can be used for real-time prediction on embedded platforms.

[0133] The above embodiment uses a causal convolution artificial neural network to extract the motion state encoding characteristics of the vehicle, which is different from the traditional RNN. The length of the historical sequence and the length of the predicted sequence can be flexibly adjusted, the model calculation has parallel characteristics, and only one forward inference is needed to complete the prediction of multiple time steps of time series. Moreover, compared with RNN or LSTM, the causal convolution artificial neural network is more lightweight, occupies less memory, and is beneficial to engineering application. The above embodiment models the vehicle dynamics completely based on data characteristics, maintains high accuracy, avoids the problem of complex and inaccurate physical modeling, and has the characteristics of low cost and high efficiency.

[0134] The embodiment also provides an electronic device, comprising: at least one processor; and a memory connected with the at least one processor in communication. The memory stores a computer program capable of being executed by the at least one processor, and the computer program, when executed by the at least one processor, is configured to cause the electronic device to perform the method according to the embodiment of the present disclosure.

[0135] The embodiment also provides a non-transitory computer readable storage medium storing a computer program, wherein the computer program, when executed by a processor of a computer, is configured to cause the computer to perform the method according to the embodiment of the present disclosure.

[0136] Reference Figure 11 The structure block diagram of the electronic device 1100 which can be a server or a client of the present disclosure will now be described, which is an example of a hardware device that can be applied to aspects of the present disclosure. The electronic device is intended to represent a wide variety of digital electronic computer devices such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computer devices. The electronic device can also represent a wide variety of mobile devices such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the present disclosure described and / or claimed in this document.

[0137] As shown in Figure 11 The electronic device 1100 includes a computing unit 1101 that can perform various appropriate actions and processes in accordance with a computer program stored in a read-only memory (ROM) 1102 or a computer program loaded from a storage unit 1108 into a random access memory (RAM) 1103. In the RAM 1103, various programs and data required for the operation of the device 1100 can also be stored. The computing unit 1101, the ROM 1102, and the RAM 1103 are connected to each other through a bus 1104. An input / output (I / O) interface 1105 is also connected to the bus 1104.

[0138] The various components in the electronic device 1100 are connected to the I / O interface 1105, including an input unit 1106, an output unit 1107, a storage unit 1108, and a communication unit 1109. The input unit 1106 can be any type of device capable of inputting information to the electronic device 1100, and can receive inputted digital or character information, and generate key signal inputs related to user settings and / or function controls of the electronic device. The output unit 1107 can be any type of device capable of presenting information, and can include, but is not limited to, a display, a speaker, a video / audio output terminal, a vibrator, and / or a printer. The storage unit 1108 can include, but is not limited to, a magnetic disk, an optical disk. The communication unit 1109 allows the electronic device 1100 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks, and can include, but is not limited to, a modem, a network card, an infrared communication device, and / or a wireless communication transceiver, such as a Bluetooth device, a WiFi device, a WiMax device, a cellular communication device, and / or the like.

[0139] The computing unit 1101 can be various general and / or special purpose processing components having processing and computing capabilities. Some examples of the computing unit 1101 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing units, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The computing unit 1101 performs various methods and processes described above. For example, in some embodiments, the above-described methods (e.g. Figure 1 The illustrated methods can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 1108. In some embodiments, portions or all of the computer program can be loaded and / or installed onto the electronic device 1100 via the ROM 1102 and / or the communication unit 1109. In some embodiments, the computing unit 1101 can be configured to perform the above-described methods by way of other any appropriate mechanism, such as by way of firmware.

[0140] Program code for carrying out methods of the present disclosure can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the program code, when executed by the processor or controller, produces the functions / operations specified in the flowcharts and / or the block diagrams. The program code can be executed entirely on a machine, partially on a machine, partially on a machine as a stand-alone software package, partially on a machine and partially on a remote machine or entirely on a remote machine or server.

[0141] In the context of this disclosure, a machine-readable medium can be a tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or other propulsive system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium will include one or more lines of electrical connections, portable computer disks, hard disk drives, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or Flash memory), optical fibers, portable compact disc read-only memories (CD-ROMs), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0142] As used in this disclosure, the terms "machine-readable medium" and "computer- readable medium" refer to any computer program product, apparatus and / or device (e.g., magnetic discs, optical disks, memory, Programmable Logic Devices (PLDs)) used to provide machine instructions and / or data to a programmable processor, including a machine-readable medium that receives machine instructions as a machine-readable signal. The term "machine-readable signal" refers to any signal that can be used to provide machine instructions and / or data to a programmable processor.

[0143] To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse) by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0144] The systems and techniques described here can be implemented in a computing system that includes a back end component, e.g., as a data server, or that includes a middleware component, e.g., an application server, or that includes a front end component, e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here, or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication, e.g., a communication network. Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0145] The computer system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.

[0146] It is to be understood that the specific embodiments described herein are meant to be illustrative only and not limiting as to the scope of the application. Any and all embodiments derived from the embodiments provided herein that are within the scope of the present application are deemed to be within the scope of the present application.

[0147] It is apparent that the illustrated embodiments are only examples of the application and are not intended to limit the scope of the application. One of ordinary skill in the art can, when hazards of brevity, modify certain aspects of the application as they can occur to those reasonably skilled in the art, without having inventive effort, and it is therefore intended that such modifications are to be considered within the scope of the application. It is also to be understood that the application is not limited to the specific embodiments, and that the specific embodiments are included to aid in understanding the overall technology. Therefore, the application as claimed hereinafter is intended to define the scope of the application and can be practiced otherwise than is specifically described herein.

[0148] The word "example" is used herein to mean serving as an example, instance, or illustration. Any aspect or embodiment described herein as "example" is not necessarily to be construed as preferred or advantageous over other aspects or embodiments. The various examples merely accomplish variations on similar devices and / or methods. Various modifications of the examples as described herein can be constructed without departing from the scope of the application. It is to be understood that the terminology "some" refers to one or more than one, and the terminology "another" refers to one or more than one of similar feature(s).

[0149] The above-described embodiments are merely illustrative of several embodiments of the present application, which are described in more detail and in a specific manner, but should not be construed as limiting the scope of protection of the present application. It should be noted that, for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the scope of protection of the present application. Therefore, the scope of protection of the present application should be subject to the appended claims.

Claims

1. A method for predicting the motion state response of a vehicle, characterized in that, include: Based on the vehicle's historical control quantities before the current moment and the vehicle's historical motion state response quantities generated in response to the historical control quantities, the motion state coding features of the vehicle at the current moment are extracted. The desired control quantity of the vehicle at the current moment is feature-encoded and mapped to the same feature dimension as the motion state encoding feature, which serves as the desired control quantity encoding feature of the vehicle after the current moment, wherein the desired control quantity is used to control the motion state response quantity of the vehicle after the current moment. The motion state coding features and the desired control quantity coding features are fused to obtain the fused features; The fused features are decoded to obtain the predicted motion state response of the vehicle after the current moment; The predicted motion state response is decoded using a decoder based on an artificial neural network; the decoder includes an input layer, a hidden layer, and an output layer. Decoding the fused features to obtain the predicted motion state response of the vehicle after the current moment includes: The fusion features are input from the input layer; Based on the input data of the node corresponding to time t in the input layer, the output result of the node corresponding to time t-1 in the output layer, and the hidden state of the node corresponding to time t-1 in the hidden layer, the hidden state of the node corresponding to time t-1 in the hidden layer is obtained. Based on the hidden state of the node corresponding to time t in the hidden layer and the output result of the node corresponding to time t-1 in the output layer, the output result of the node corresponding to time t in the output layer is obtained as the predicted motion state response of the vehicle at time t.

2. The method according to claim 1, characterized in that, Before extracting the motion state encoding features of the vehicle at the current moment based on the vehicle's historical control quantities before the current moment and the historical motion state response quantities generated by the vehicle in response to the historical control quantities, the method further includes: sampling the historical control quantities and the historical motion state response quantities to obtain a historical control quantity sequence and a historical motion state response quantity sequence; and fusing the historical control quantity sequence and the historical motion state response quantity sequence according to the sampling time corresponding to each element in the historical control quantity sequence and the historical motion state response quantity sequence to obtain a fused sequence. Extracting the motion state coding features of the vehicle at the current moment based on the vehicle's historical control quantities prior to the current moment and the historical motion state response quantities generated by the vehicle in response to the historical control quantities includes: extracting the motion state coding features of the vehicle at the current moment based on the fusion sequence.

3. The method according to claim 2, characterized in that, The motion state encoding features are extracted using a first encoder based on an artificial neural network. The first encoder includes an input layer, a hidden layer, and an output layer. Specifically, extracting the motion state encoding features of the vehicle at the current moment based on the fusion sequence includes: The fusion sequence is input from the input layer; For each hidden layer: When the layer above the current hidden layer is the input layer, the hidden state of the node corresponding to time t in the current hidden layer is obtained based on the input data of the n+1 nodes corresponding to time tn in the input layer and the output result of the node corresponding to time t-1 in the output layer. When the layer above the current hidden layer is a hidden layer, based on the hidden states of the n+1 nodes corresponding to time t to time tn in the hidden layer above, obtain the hidden state of the node corresponding to time t in the current hidden layer. When the layer below the current hidden layer is the output layer, the node corresponding to time t in the output layer processes the hidden state of the current hidden layer and obtains the output result, which serves as the motion state encoding feature of the vehicle at time t.

4. The method according to claim 2, characterized in that, The motion state encoding features are extracted using a first encoder based on an artificial neural network. The first encoder includes an input layer, a hidden layer, and an output layer. Specifically, extracting the motion state encoding features of the vehicle at the current moment based on the fusion sequence includes: The fusion sequence is input from the input layer; For each hidden layer: When the layer above the current hidden layer is the input layer, the hidden state of the node corresponding to time t in the current hidden layer is obtained based on the input data of the n+1 nodes corresponding to time tn in the input layer and the output result of the node corresponding to time t-1 in the output layer. When the layer above the current hidden layer is a hidden layer, n+1 hidden states are selected from the hidden states of the nodes corresponding to the time t to the time tn in the hidden layer above, according to equal time intervals. And based on the selected n+1 hidden states, the hidden state of the node corresponding to the time t in the current hidden layer is obtained. The time interval increases with the number of hidden layers. When the layer below the current hidden layer is the output layer, the node corresponding to time t in the output layer processes the hidden state of the current hidden layer and obtains the output result, which serves as the motion state encoding feature of the vehicle at time t.

5. The method according to claim 3 or 4, characterized in that, After the node corresponding to time t in the output layer processes the hidden state of the current hidden layer and obtains the output result as the motion state encoding feature of the vehicle at time t, the method further includes: If the output layer includes nodes corresponding to future times, then the output results of the nodes corresponding to future times are removed from the output results of the output layer; and The motion state encoding features are defined by extracting a preset number of output results from the output layer between the current time and historical time.

6. The method according to claim 1, characterized in that, The desired control quantity encoding features are extracted using a second encoder based on an artificial neural network. The second encoder includes an input layer, a hidden layer, and an output layer. The desired control quantity is feature-encoded and mapped to the same feature dimension as the motion state encoding features. The desired control quantity encoding features for the vehicle after the current moment include: The desired control quantity is sampled to obtain a sequence of desired control quantities; The desired control quantity sequence is input from the input layer; For each hidden layer: When the layer above the current hidden layer is the input layer, the hidden state of each node in the current hidden layer is obtained based on the input data of all nodes in the input layer. When the layer above the current hidden layer is a hidden layer, obtain the hidden state of each node in the current hidden layer based on the hidden state of all nodes in the hidden layer above. When the layer below the current hidden layer is the output layer, the output layer processes the hidden state of each node in the current hidden layer and obtains the output result, which is used as the expected control quantity encoding feature. The number of nodes in the output layer is the same as the number of feature dimensions of the motion state encoding feature.

7. The method according to claim 1, characterized in that, By fusing the motion state encoding features and the desired control quantity encoding features, the fused features include: The motion state encoding features and the desired control quantity encoding features are fused using a parallel strategy to obtain the fused features.

8. A predictive device for vehicle motion state response, characterized in that... include: The first encoder is used to extract the motion state coding features of the vehicle at the current moment based on the historical control quantities of the vehicle before the current moment and the historical motion state response quantities generated by the vehicle in response to the historical control quantities. The second encoder is used to encode the desired control quantity of the vehicle at the current moment and map it to the same feature dimension as the motion state encoding feature, as the desired control quantity encoding feature of the vehicle after the current moment, wherein the desired control quantity is used to control the motion state response quantity of the vehicle after the current moment. The fusion module is used to fuse the motion state coding features and the desired control quantity coding features to obtain fused features; A decoder is used to decode the fused features to obtain the predicted motion state response of the vehicle after the current moment; The decoder is an artificial neural network-based decoder; the decoder includes an input layer, a hidden layer, and an output layer. The decoder is used for: The fusion features are input from the input layer; Based on the input data of the node corresponding to time t in the input layer, the output result of the node corresponding to time t-1 in the output layer, and the hidden state of the node corresponding to time t-1 in the hidden layer, the hidden state of the node corresponding to time t-1 in the hidden layer is obtained. Based on the hidden state of the node corresponding to time t in the hidden layer and the output result of the node corresponding to time t-1 in the output layer, the output result of the node corresponding to time t in the output layer is obtained as the predicted motion state response of the vehicle at time t.

9. An electronic device, comprising: processor; And a memory for storing a program, characterized in that the program includes instructions that, when executed by the processor, cause the processor to perform the method according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 7.

Images