CAE‐GAN: A hybrid model for vehicle trajectory prediction

Abstract

Trajectory prediction of surrounding vehicles is a crucial capability of intelligent driving vehicles. In a scene, a vehicle and its surrounding vehicles constitute an integral system, and the vehicle's future motion trajectory is affected by the actions of surrounding vehicles. The influencing mode and degree are hidden in the relevant historical information of the vehicle and its neighbour vehicle. The existing trajectory prediction methods either do not consider the confidence of the predicted trajectory, or the accuracy requirement is ignored when considering the confidence of the predicted trajectory. In order to address this problem, a mixed Conditional AutoEncoder Generative Adversarial Network (CAE‐GAN) model based on the multi‐loss function is proposed. The proposed model uses the encoder–decoder structure with a convolutional social pool to extract general features. The generative adversarial networks (GANs) are used to extract the confidence features of the generated trajectories, which enables the proposed model to generate trajectories that are close to the real trajectories. In addition, a classifier structure based on an LSTM network is added to output the probability that the predicted trajectory belongs to a particular lateral maneuver so that the generated trajectory lateral maneuver of the model is consistent with the real trajectory lateral maneuver. The proposed model is evaluated using the publicly available NGSIM US‐101 and I‐80 datasets, and results show that the accuracy of the proposed model is higher than that of the existing methods. The proposed model achieves an average accuracy improvement of 16.34% in comparison to the most advanced existing models.

Deep generative models for vehicle speed trajectories

“…A GAN model is used in Reference 80 to generate vehicle trajectories in intersections so they could model vehicle behaviors relative to other vehicles. An LSTM‐based GAN is used in Reference 78 to predict the trajectory of surrounding vehicles.…”

“…The most recent research in vehicle trajectory generation includes exploring different combinations of DL models to generate the trajectories. [77][78][79][80] In Reference 77 a combination of an RNN embedding and a GAN is used to generate the trajectory of vehicle locations and thus the movement pattern of the vehicle. In Reference 79 an LSTM-based GAN is used to generate driving patterns.…”

Traffic Agents Trajectory Prediction Based on Spatial–Temporal Interaction Attention

“…Experimental results show that the GSTA improves the pedestrian trajectory prediction accuracy. Based on GAN, a mixed Conditional Auto Encoder Generative Adversarial Network (CAE-GAN) [ 26 ] model using the multi-loss function is proposed. The GAN in the CAE-GAN is used to extract features of the generated trajectories and generate trajectories that are close to the real trajectories.…”

A cognition‐inspired trajectory prediction method for vehicles in interactive scenarios

“…To predict vehicle trajectories in interactive scenarios, many scholars focus on the vehicles' behavioral characteristics and attempt to figure out the common rules behind these behaviors based on the advantages of DL algorithms. For example, convolutional neural networks (CNN) [8] work well at learning features in the picture, so they are usually used to embed environmental features from pictures or grids [9,10]; long short-term networks (LSTM) [11] are good at dealing with sequential information, so they are used to abstract the features of vehicle trajectories [12,13]; graph neural networks (GNN) [14] work well at learning interaction relationships, so they are used to model the interaction between vehicles [15,16]; fully connected layers (FC) can fit complicated functions directly, and thus they often work as the output layers which get desired outputs from hidden layers [17,18]. Besides, the autoencoder, which contains an encoder and a decoder, can abstract primary features of the input information, so it is frequently used to learn latent states from historical trajectories in the encoder and predict trajectories in the future in the decoder [19,20].…”

“…Besides, the autoencoder, which contains an encoder and a decoder, can abstract primary features of the input information, so it is frequently used to learn latent states from historical trajectories in the encoder and predict trajectories in the future in the decoder [19,20]. Generative Adversarial Networks (GAN), which are trained by the gaming of the generator and discriminator, can also learn the hidden features of the input information and be used to predict trajectories [12,21]. Briefly, DL-based methods learn the models from massive data and search for possible functions in a large space.…”