A method for predicting uncertain trajectories of traffic participants based on condition fusion

Abstract

The present invention provides a method for predicting the uncertain trajectory of traffic participants based on conditional fusion, which relates to the field of trajectory prediction of autonomous driving vehicles. The method collects the status information and lane information of each participant in the traffic scene for a predetermined period of time in the past and packages them into an embedded vector sequence, maps the embedded vector sequence into a potential array, inputs the potential array and the embedded vector sequence into the encoding network to obtain high-dimensional deep representation information; inputs the high-dimensional deep representation information into the diffusion model for training; uses the trained diffusion model to generate the predicted trajectory of the target, and obtains the trajectory prediction result considering uncertainty through multiple sampling. This solution integrates the high-dimensional representation of scene information by the deep neural network and the probabilistic characteristics generated by multiple sampling of the diffusion model to characterize the various future trajectory branches that may appear in the vehicle under the same historical conditions, providing a reliable basis for autonomous driving decision-making and path planning in complex traffic scenarios.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving vehicle trajectory prediction, and in particular to a method for predicting the uncertain trajectories of traffic participants based on conditional fusion. Background Art

[0002] Trajectory prediction is a critical task in the fields of transportation and autonomous driving. It aims to infer the potential future positions and trajectories of traffic participants (such as vehicles, pedestrians, and cyclists) based on historical observations and environmental information. High-precision trajectory prediction is a crucial prerequisite for safe and efficient decision-making (e.g., planning, collision avoidance, speed control, etc.).

[0003] Prediction methods based on physical models, such as Kalman filtering, can model and predict future trajectories in the short term by using the vehicle's historical short-term state information. However, motion prediction results in the long term are unreliable.

[0004] In recent years, data-driven approaches have become increasingly mainstream. Collecting large datasets to train neural networks to capture behavioral characteristics in traffic environments for trajectory prediction has significantly improved prediction accuracy, particularly in modeling interactions between traffic participants within a scene. The application of models such as RNNs, LSTMs, Transformers, and graph neural networks to time series data has enabled significant progress in analyzing dynamic interactions over long timescales and in complex scenarios.

[0005] In real-world traffic scenarios, the same historical characteristics can often lead to multiple future directions, such as continuing straight or merging. This makes it difficult to fully model future uncertainty using a single predicted trajectory. While many existing solutions can predict multimodal position coordinates to a certain extent, they struggle to represent the diverse movement patterns of various traffic participants, posing a potential safety threat to autonomous driving decision-making and planning. Summary of the Invention

[0006] Purpose of the invention: To propose a method for predicting the uncertain trajectories of traffic participants based on conditional fusion to solve the above-mentioned problems existing in the prior art.

[0007] The method for predicting uncertain trajectories of traffic participants based on conditional fusion proposed in this invention comprises the following steps:

[0008] Collect status information and lane information of each participant in the traffic scene over the past predetermined period of time;

[0009] The state information and lane information are packaged into an embedding vector sequence X, the embedding vector sequence X is mapped to a latent array Z, and the latent array Z and the embedding vector sequence X are input into an encoding network to obtain high-dimensional deep representation information C; the encoding network is stacked with several layers of encoders, each of which includes a cross-attention module and a latent self-attention module;

[0010] Select a diffusion model, input the high-dimensional depth representation information C into it, and complete the training of the diffusion model;

[0011] The trained diffusion model is used to generate the predicted trajectory of the target, and the trajectory prediction result considering uncertainty is obtained through multiple sampling.

[0012] In a further embodiment, the status information is represented as follows:

[0013] X traj_i =[x,y,vx,vy,ax,ay]

[0014] Where, X traj_i represents the i-th traffic participant in the scene, x, y, vx, vy, ax, ay represent the coordinates, velocity, and acceleration information of the participant in the horizontal and vertical directions respectively;

[0015] Lane information is represented as follows:

[0016] X lane_i =[x lane ,y lane ,point]

[0017] Where, X lane_i Indicates the information of the i-th lane line, x lane ,y lane ,point respectively represents the horizontal coordinate, vertical coordinate and direction of the lane line sampling point.

[0018] In a further embodiment, the state information and lane information are packaged into an embedding vector sequence X, specifically including:

[0019] Calculate the embedded features of the vehicle, neighboring vehicles, and lane information:

[0020]

[0021] Where, e ego 、e nbrs,i 、,e lane,i Represent the embedded features of the vehicle, neighboring vehicle, and lane information respectively; W ego 、W nbr 、W lane is the weight coefficient of each embedding layer; b ego 、bnbr 、b lane is the bias; each embedding feature dimension is d x ;

[0022] Combine the embedded features of the vehicle, neighboring vehicles, and lane information into an embedded vector sequence X:

[0023] X={e ego ,e lane,1 ,…,e nbr,1 ,e nbr,2 ,…,e nbr,n}.

[0024] In a further embodiment, the embedding vector sequence X is mapped to a potential array Z, and the potential array Z and the embedding vector sequence X are input into the encoding network to obtain high-dimensional deep representation information C, which specifically includes:

[0025] Initialize the potential array

[0026]

[0027] Where M is the size of the potential array, d z is the dimension of the latent vector;

[0028] The underlying array that will be initialized The embedded vector sequence X is input into the encoding network, and the high-order representation feature Z of the multi-time historical information in the traffic scene is obtained through the n-layer stacking of the encoding network. n , and obtain the conditional information C used to input the diffusion model through nonlinear mapping.

[0029] In a further embodiment, the potential array to be initialized And the embedding vector sequence X is input into the encoding network, and the encoding process of the lth layer is as follows:

[0030] The cross attention mechanism converts the latent vector of the l-1 layer As the query vector, the input embedding vector sequence X is used as the key and value, and the attention mechanism is used to allow the latent vector to absorb multimodal data information to obtain the output of the first layer.

[0031] In a further embodiment, the cross attention output result of the i-th layer is obtained by the cross attention mechanism and the residual connection

[0032]

[0033] Where, represents the query vector of the mth latent vector in the crisscross attention layer i, Represent the key and value vectors of the input i-th token in the i-th layer respectively; is a trainable coefficient matrix, and layernorm(*) is a layer normalization operation.

[0034] In a further embodiment, self-attention is performed between latent vectors to allow the latent vectors to be fused with each other to form the final output encoding result

[0035]

[0036] Where, The query, key, and value of the latent vector in the self-attention mechanism respectively.

[0037] In a further embodiment, after n layers of stacking, the final latent vectors are concatenated to obtain the high-order representation feature Z of the multi-time historical information in the traffic scene. n , and obtain the conditional information C for inputting the diffusion model through nonlinear mapping:

[0038]

[0039] C=MLP(Z n )

[0040] Where MLP(*) is a nonlinear mapping operation.

[0041] In a further embodiment, the high-dimensional depth representation information C is input into an open-source diffusion model to complete the training of the diffusion model, which specifically includes:

[0042] Noise adding stage: the target future trajectory Y 0 Gradually add Gaussian noise ε of a predetermined magnitude for forward noise addition. After k rounds of iteration, the noisy trajectory Y is obtained. k ;

[0043] Training phase: In each round of iteration, let the denoising network ∈ θ At the same time, read the noisy track Y k , high-dimensional depth representation information C and the current iteration number k, output prediction noise Predicting noise by calculation The mean square error with the actual Gaussian noise ε is used to train the network through back propagation;

[0044] Inference stage: Initialize pure Gaussian noise Input to the trained denoising network ∈ θ In , the predicted trajectory is obtained by reverse denoising.

[0045] In a further embodiment, the reasoning stage specifically includes the following process:

[0046] Call the denoising network:

[0047]

[0048] Where, Indicates the model's estimate of the noise during iteration, and m is the current number of steps;

[0049] Backward Diffusion Update:

[0050]

[0051] Where, represents the noisy trajectory of the current step, is the noisy trajectory of the m-1th step after update, β m is the noise coefficient, α m In the inference process, it is used to determine the variance scaling, n m ~Ν(0,Ι) means that random sampling can still be performed during the denoising process;

[0052] The recursive direction is diffused until m-1=0, and the final future prediction trajectory is obtained;

[0053] By repeatedly sampling different initial noise and n at each step m And perform the reverse diffusion update to generate different prediction trajectory results.

[0054] In addition, the present invention also proposes an electronic device, which includes: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the above-mentioned conditional fusion-based uncertain trajectory prediction method for traffic participants is implemented.

[0055] In addition, the present invention also proposes a computer-readable storage medium, which stores at least one executable instruction. When the executable instruction is run on an electronic device, the electronic device executes the above-mentioned conditional fusion-based uncertainty trajectory prediction method for traffic participants.

[0056] Compared with the prior art, the present invention has at least the following beneficial effects:

[0057] (1) The Perceiver structure (latent array + cross attention + latent self-attention) is used to efficiently represent the state of the vehicle and neighboring vehicles at several moments in the past, rather than a single RNN or other conventional time series network structure. RNN networks may encounter the problem of vanishing gradients when processing data in long time series; and when the self-attention mechanism is used on a large scale, N token sequences will be generated. 2This solution can aggregate the time series information of a large number of moments and multiple vehicles to form a concentrated and expressive latent vector. Moreover, through a fixed-size latent array (M<<N), it can achieve controllable computational complexity and strong expressiveness.

[0058] (2) Combine the diffusion model to generate future trajectories. This solution can capture various behavior patterns in driving scenarios through large-scale data training, and obtain multiple uncertain trajectory branches from the sampling of CNG random noise. Conventional multimodal trajectory prediction only achieves rough multimodal branch network division through discrete intent classification, which cannot capture more possibilities. This solution can generate multiple trajectory distributions in a continuous probability space through random initialization and randomness between steps.

[0059] (3) Injecting Perceiver conditions into the denoising network strongly correlates trajectory generation with historical scenarios. In conventional diffusion model denoising networks, deep contextual integration of temporal models is rarely performed. This method explicitly incorporates the historical features of multiple vehicles into the denoising network, generating conditions based on driving scenarios. This allows the denoising network to combine historical information at each step for more accurate denoising and restoration. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a schematic diagram of the model structure constructed based on the prediction method proposed in the embodiment of the present invention.

[0061] Figure 2 Schematic diagram of a traffic participant uncertainty trajectory prediction system proposed in an embodiment of the present invention. DETAILED DESCRIPTION

[0062] In the following description, numerous specific details are provided to provide a more thorough understanding of the present invention. However, it will be apparent to those skilled in the art that the present invention may be practiced without one or more of these details. In other instances, certain technical features well known in the art have not been described to avoid confusion with the present invention.

[0063] To address the shortcomings of existing technologies, the embodiments of the present invention are based on the idea of ​​"uncertain trajectory prediction", integrating the high-dimensional representation of scene information by deep neural networks and the probabilistic characteristics generated by multiple sampling of diffusion models, focusing on depicting the various future trajectory branches that may appear for vehicles under the same historical conditions. The distribution of future trajectories is learned through end-to-end training. During inference, only multiple sampling of the initial noise is required to obtain prediction results that are closely related to the historical environment but different from each other, thereby improving the ability to express uncertainty in real scenarios and providing a reliable foundation for decision-making and path planning in autonomous driving and complex traffic scenarios.

[0064] This embodiment discloses a method for predicting the uncertain trajectory of traffic participants based on conditional fusion, and its data flow is shown in FIG. Figure 1 .

[0065] Here are the steps:

[0066] S1. Collect the status information of each participant in the traffic scene in the past 3 seconds, including location coordinates, speed, acceleration, and traffic road line information:

[0067] X traj_i =[x,y,vx,vy,ax,ay]

[0068] X lane_i =[x lane ,y lane ,point]

[0069] Where, X traj_i represents the i-th traffic participant in the scene, and x, y, vx, vy, ax, and ay represent the coordinates, speed, and acceleration information of the participant in the horizontal and vertical directions respectively. lane_i Represents the information of the lane line at position i, x lane ,y lane ,point respectively represent the coordinates and orientation of the lane line sampling point.

[0070] S2 packages the information obtained in S1 into a sequence and maps it to the potential array Z of the Perciever encoder through input token mapping. After undergoing a multi-layer stacking process of cross attention and potential self-attention, it finally obtains high-dimensional deep representation information C for subsequent multimodal trajectory generation.

[0071] S3, based on the encoded information C obtained by S2 encoding, uses the diffusion model to generate the predicted trajectory of the target through denoising, and can obtain the trajectory prediction result considering uncertainty through multiple sampling.

[0072] As a preferred embodiment, in step S1, the vehicle's trajectory information and speed acceleration information can be obtained by loading on-board sensors, a vehicle bus, and a vehicle body positioning module on the vehicle. The state parameters of surrounding vehicles can be sensed and tracked in real time by using millimeter-wave radar and cameras. The lane line pixels are identified and converted into road coordinates by combining the front camera with the lane line detection algorithm.

[0073] As a preferred embodiment, the diffusion model uses a mainstream open-source diffusion model, such as Stable Diffusion developed by StabilityAI. It supports functions such as text-to-image generation, image restoration, and super-resolution. Its core is based on the DDPM framework, optimized for generation efficiency, and can run on consumer-grade graphics cards.

[0074] For another example, you can choose DDPM (Denoising Diffusion Probabilistic Models), which supports custom noise schedulers (such as linear and cosine noise attenuation).

[0075] For example, you can use the Diffusers library (Hugging Face). This is an open source library released by Hugging Face that integrates multiple diffusion models (such as Stable Diffusion and DALL·E mini) and provides a unified API interface.

[0076] As a preferred embodiment, the encoding process of the mapping and model in step S2 is disclosed:

[0077] S2.1. Embed and merge the data obtained in S1:

[0078]

[0079] Where e ego ,e nbrs,i ,e lane,i Represents the embedded feature representation of the vehicle, neighboring vehicles, and lane information, W ego ,W nbr ,W lane is the weight coefficient of the embedding layer. b ego ,b nbr ,b lane is the bias. All embedded features are integrated into a set X, where each embedded feature dimension is d x .

[0080] S2.2. Initialize the potential array

[0081]

[0082] Where M is the size of the potential array, d z is the dimension of the latent vector, It can be randomly initialized at the beginning of training and used as a learnable parameter.

[0083] S2.3. Potential array to be initialized The embedding vector sequence X is input into the encoding network. The encoder consists of n stacked layers, each of which consists of two steps: cross attention and potential self-attention. After a multi-layer stacking process of cross attention and potential self-attention mechanisms, high-dimensional deep features that integrate the modal information in the historical time domain of the scene are finally obtained.

[0084] As a preferred embodiment, a feasible solution detail of step S2.3 is disclosed:

[0085] S2.3.1, take the encoding process of layer l as an example: First, the cross attention mechanism converts the latent vector of layer l-1 into As the query vector, the input token sequence X is used as the key and value, and the attention mechanism is used to allow the latent vector to absorb multimodal data information to obtain the output of the first layer. The specific process is as follows:

[0086] First, the cross attention mechanism and residual connection are used to obtain the cross attention output result of the i-th layer

[0087]

[0088] Where, represents the query vector of the mth latent vector in the crisscross attention layer i, Represents the key and value vectors of the input i-th token in the i-th layer respectively. is a trainable coefficient matrix, and layernorm is a layer normalization operation.

[0089] Next, we perform self-attention between the latent vectors to allow them to fuse with each other to form a higher-order latent representation:

[0090]

[0091] in The query, key, and value of the latent vector in the self-attention mechanism respectively. The final output encoding result of this layer.

[0092] S2.3.2. After stacking n layers, the final latent vectors are concatenated to obtain the high-order representation feature Z of the multi-time historical information in the traffic scene. n , and obtain the conditional information C for inputting the diffusion model through nonlinear mapping:

[0093]

[0094] C=MLP(Z n )

[0095] Where MLP is a nonlinear mapping operation.

[0096] As a preferred embodiment, details of a feasible solution for step S3 are disclosed:

[0097] S3.1, according to the conventional diffusion model training method, first add noise: during the training process, the target future trajectory Y 0Gradually add a small amount of Gaussian noise ε for forward noise addition. After k rounds of iteration, the trajectory becomes almost pure noise Y k .

[0098] S3.2. According to the conventional diffusion model training method, the denoising network is trained: at each diffusion step k, let the denoising network ∈ θ Simultaneously read the noisy track Y k , Perceiver encoder condition C and current time k, output prediction of noise amount

[0099] Predicting noise by calculation The mean square error between the noise and the actual noise ε is used to train the network through back propagation.

[0100] S3.3, inference stage, there is no longer a real target future trajectory Y 0 , initialize pure Gaussian noise Input to the trained denoising network ∈ θ In , the predicted trajectory is obtained by reverse denoising.

[0101] S3.3.1. Call the denoising network:

[0102]

[0103] In the formula It represents the noise estimation of the model in the iteration, and m is the current number of steps.

[0104] S3.3.2. Backward Diffusion Update:

[0105]

[0106] Where, represents the noisy trajectory of the current step, is the noisy trajectory of the m-1th step after update, β m is the noise coefficient, α m In the inference process, it is used to determine the variance scaling, n m ~Ν(0,Ι) means that random sampling can still be performed during the denoising process, making the result more random.

[0107] S3.3.3. Recursively diffuse the direction in the above manner until m-1=0 to obtain the final future prediction trajectory.

[0108] S3.3.4. By repeatedly sampling different initial noises and n at each step m , and perform the above reverse denoising process to generate different predicted trajectory results.

[0109] As a preferred embodiment, see Figure 2 As shown, a traffic participant uncertainty trajectory prediction system 400 is disclosed, and the trajectory prediction system 400 at least includes an acquisition module 401, a first data processing module 402, a second data processing module 403, a model training module 404, and a prediction result output module 405.

[0110] The collection module 401 is used to collect status information and lane information of each participant in the traffic scene over a predetermined period of time in the past.

[0111] The first data processing module 402 is used to package the collected state information and lane information into an embedding vector sequence X.

[0112] The second data processing module 403 embeds an encoding network. The encoding network consists of several stacked encoder layers, each of which includes a cross-attention module and a latent self-attention module. The second data processing module 403 maps the embedding vector sequence X to a latent array Z. The latent array Z and the embedding vector sequence X are input into the encoding network to obtain a high-dimensional deep representation C.

[0113] The model training module 404 is used to input the high-dimensional depth representation information C into the open source diffusion model to complete the training of the diffusion model.

[0114] The trained diffusion model is used to generate a predicted trajectory of the target, and a trajectory prediction result that takes uncertainty into account is obtained through multiple samplings and outputted through the prediction result output module 405 .

[0115] As a preferred embodiment, an electronic device is disclosed, comprising a processor, a memory, a communication interface, and a communication bus. The processor, the memory, and the communication interface communicate with each other via the communication bus. The memory encapsulates at least one computer-readable storage medium, which stores at least one executable instruction. When the computer instruction or computer program is loaded or executed on the electronic device, the processes or functions described in the embodiments of this application are fully or partially generated, which will not be described in detail here.

[0116] The electronic device may also communicate with one or more external devices (e.g., keyboards, pointing devices, Bluetooth devices, etc.), one or more devices that enable a user to interact with the electronic device, and / or any device that enables the electronic device to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface. Furthermore, the electronic device may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter. The network adapter communicates with other modules of the electronic device via a bus. It should be understood that, although not shown in the figures, other hardware and / or software modules may be used in conjunction with the electronic device, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0117] It should be noted that the above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center via a wired (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0118] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0119] As described above, although the present invention has been shown and described with reference to specific preferred embodiments, it should not be construed as limiting the present invention itself. Various changes may be made to it in form and detail without departing from the spirit and scope of the present invention as defined in the appended claims.

Claims

1. A method for predicting uncertain trajectories of traffic participants based on conditional fusion, characterized in that: The steps include: Collect status information and lane information of each participant in the traffic scene over the past predetermined period of time; The state information and lane information are packaged into an embedding vector sequence X, the embedding vector sequence X is mapped to a latent array Z, and the latent array Z and the embedding vector sequence X are input into an encoding network to obtain high-dimensional deep representation information C; the encoding network is stacked with several layers of encoders, each of which includes a cross-attention module and a latent self-attention module; Inputting the high-dimensional depth representation information C into an open-source diffusion model to complete the training of the diffusion model; The trained diffusion model is used to generate the predicted trajectory of the target, and the trajectory prediction result considering uncertainty is obtained through multiple sampling.

2. The method for predicting uncertain trajectories of traffic participants based on conditional fusion according to claim 1 is characterized in that: The status information is expressed in the following form: Where, represents the i-th traffic participant in the scene, Respectively represent the coordinates, velocity, and acceleration information of the participant in the horizontal and vertical directions; The lane information is expressed as follows: Where, Represents the information of the i-th lane line, They represent the horizontal coordinate, vertical coordinate, and direction of the lane line sampling point respectively.

3. The method for predicting uncertain trajectories of traffic participants based on conditional fusion according to claim 2 is characterized in that: The state information and lane information are packaged into an embedding vector sequence X, specifically including: Calculate the embedded features of the vehicle, neighboring vehicles, and lane information: Where, 、 、 Represent the embedded features of the vehicle, neighboring vehicle, and lane information respectively; 、 、 is the weight coefficient of each embedding layer; 、 、 is the bias; the size of each embedded feature dimension is ; Combine the embedded features of the vehicle, neighboring vehicles, and lane information into an embedded vector sequence X: 。 4. The method for predicting uncertain trajectories of traffic participants based on conditional fusion according to claim 1, characterized in that: Map the embedded vector sequence X to a potential array Z, input the potential array Z and the embedded vector sequence X into the encoding network to obtain high-dimensional deep representation information C, specifically including: Initialize the potential array : Where M is the size of the potential array, is the dimension of the latent vector; The underlying array that will be initialized And the embedding vector sequence X is input into the encoding network, and the high-order representation features of multi-time historical information in the traffic scene are obtained through the n-layer stacking of the encoding network , and obtain the conditional information C used to input the diffusion model through nonlinear mapping.

5. The method for predicting uncertain trajectories of traffic participants based on conditional fusion according to claim 1 or 4, characterized in that: The underlying array that will be initialized And the embedding vector sequence X is input into the encoding network, and the encoding process of the lth layer is as follows: The cross attention mechanism The latent vector of the layer As the query vector, the input embedding vector sequence X is used as the key and value, and the attention mechanism is used to allow the latent vector to absorb multimodal data information to obtain the output of the first layer.

6. The method for predicting uncertain trajectories of traffic participants based on conditional fusion according to claim 5 is characterized in that: The first Cross-attention output of the layer : Where, Indicates the The latent vector in query vector in layer crisscross attention, Respectively represent the input token i in the The key and value vectors in the layer; is the trainable coefficient matrix, is the layer normalization operation.

7. The method for predicting uncertain trajectories of traffic participants based on conditional fusion according to claim 6, characterized in that: Perform self-attention between latent vectors to allow the latent vectors to fuse with each other to form the final output encoding result : Where, The query, key, and value of the latent vector in the self-attention mechanism respectively.

8. The method for predicting uncertain trajectories of traffic participants based on conditional fusion according to claim 7 is characterized in that: After n layers of stacking, the final latent vectors are concatenated to obtain high-level representation features of multi-time historical information in traffic scenarios. , and obtain the conditional information for inputting the diffusion model through nonlinear mapping : Where, It is a nonlinear mapping operation.

9. The method for predicting uncertain trajectories of traffic participants based on conditional fusion according to claim 1 or 8, characterized in that: The high-dimensional depth representation information C is input into the open-source diffusion model to complete the training of the diffusion model, which specifically includes: Noise adding stage: the target future trajectory Gradually add Gaussian noise of predetermined magnitude Perform forward noise addition and obtain the noisy trajectory after k rounds of iterations ; Training phase: In each round of iteration, the denoising network Simultaneously read the noisy track , high-dimensional depth representation information C and the current iteration number k, output prediction noise ; Predict noise by calculation With actual Gaussian noise The mean square error is used to train the network through back propagation; Inference stage: Initialize pure Gaussian noise , input to the trained denoising network In , the predicted trajectory is obtained by reverse denoising.

10. The method for predicting uncertain trajectories of traffic participants based on conditional fusion according to claim 9, characterized in that: The reasoning stage specifically includes: Call the denoising network: Where, represents the model's estimate of the noise in the iteration, is the current number of steps; Backward Diffusion Update: Where, represents the noisy trajectory of the current step, After the update The noisy trajectory of the step, is the noise factor, Used to determine the variance scaling during inference, Indicates that random sampling is still possible during the denoising process; The recursive direction spreads until =0, get the final future prediction trajectory; By repeatedly sampling different initial noise and each step And perform the reverse diffusion update to generate different prediction trajectory results.

Images