Vehicle Trajectory Prediction Research Articles

Kernel-Based Metrics Learning for Uncertain Opponent Vehicle Trajectory Prediction in Autonomous Racing

Kernel-Based Metrics Learning for Uncertain Opponent Vehicle Trajectory Prediction in Autonomous Racing

Multicolumn Self-Attention GRU Model for Intersection Vehicle Trajectory Prediction

Multicolumn Self-Attention GRU Model for Intersection Vehicle Trajectory Prediction

Advanced obstacle recognition and trajectory prediction for autonomous vehicles using 3D CNN and GRU

Abstract. Recent years have seen a rapid advancement in new energy vehicle technology, which has increased consumer demand for autonomous driving systems. However, current self-driving cars still face many technical barriers. Obstacle detection and obstacle trajectory prediction in dynamic environments is one of the main difficulties of the technique. To improve self-driving cars' capacity for route planning and decision-making in dynamic contexts, this thesis proposes an obstacle identification and trajectory prediction model that combines a gated recurrent unit (GRU) network with a three-dimensional convolutional neural network (3D CNN). The integrated framework utilises the feature recognition capability of 3D CNN and the prediction capability of GRU to accurately predict dynamic obstacles, thereby improving the performance of the model in various types of environments (e.g., visual SLAM). Experimental results based on publicly available datasets (e.g., KITTI and Argoverse) show that the proposed method has significant advantages in dynamic obstacle identification and prediction as well as overall model operation efficiency. The findings show that the integrated framework is useful and successful in advancing the creation of self-driving automobiles.

Hypersonic glide vehicle trajectory prediction based on frequency enhanced channel attention and light sampling-oriented MLP network

Hypersonic Glide Vehicles (HGVs) are advanced aircraft that can achieve extremely high speeds (generally over 5 Mach) and maneuverability within the Earth's atmosphere. HGV trajectory prediction is crucial for effective defense planning and interception strategies. In recent years, HGV trajectory prediction methods based on deep learning have the great potential to significantly enhance prediction accuracy and efficiency. However, it's still challenging to strike a balance between improving prediction performance and reducing computation costs of the deep learning trajectory prediction models. To solve this problem, we propose a new deep learning framework (FECA-LSMN) for efficient HGV trajectory prediction. The model first uses a Frequency Enhanced Channel Attention (FECA) module to facilitate the fusion of different HGV trajectory features, and then subsequently employs a Light Sampling-oriented Multi-Layer Perceptron Network (LSMN) based on simple MLP-based structures to extract long/short-term HGV trajectory features for accurate trajectory prediction. Also, we employ a new data normalization method called reversible instance normalization (RevIN) to enhance the prediction accuracy and training stability of the network. Compared to other popular trajectory prediction models based on LSTM, GRU and Transformer, our FECA-LSMN model achieves leading or comparable performance in terms of RMSE, MAE and MAPE metrics while demonstrating notably faster computation time. The ablation experiments show that the incorporation of the FECA module significantly improves the prediction performance of the network. The RevIN data normalization technique outperforms traditional min-max normalization as well.

Multimodal vehicle trajectory prediction based on intention inference with lane graph representation

Multimodal vehicle trajectory prediction based on intention inference with lane graph representation

Advancing Vehicle Trajectory Prediction: A Probabilistic Approach Using Combined Sequential Models

Abstract This paper addresses the critical need to quantify vehicle trajectory uncertainty in autonomous driving under environmental variability. We focus on predicting the posterior distribution of vehicle trajectories over a fixed horizon, given an initial state and a sequence of actions. We propose and compare three approaches: a probabilistic seq2seq model based on stochastic variational Gaussian processes, sequential Monte Carlo simulation with a single-step Gaussian process model, and a hybrid model that leverages the strengths of both methods. Each approach incorporates a baseline vehicle kinematics model to enhance stability and convergence. We evaluate these methods using a dataset generated from the CARLA simulator, assessing both point error metrics and probabilistic prediction metrics. This research introduces novel approaches to quantifying vehicle trajectory uncertainty through various uncertainty quantification techniques, with the goal of improving the safety and reliability of autonomous vehicle control systems.

PillarVTP: vehicle trajectory prediction method based on local point cloud aggregation and receptive field expansion

PillarVTP: vehicle trajectory prediction method based on local point cloud aggregation and receptive field expansion

Research on vehicle trajectory prediction methods in dense and heterogeneous urban traffic

ABSTRACT In autonomous driving, accurately predicting the trajectories of surrounding vehicles is essential, particularly in dense and heterogeneous urban traffic. We propose a graph-structured model with a category layer to efficiently forecast the target vehicle’s trajectory. The model enables flexible selection of interacting objects based on environmental interactions and extracts spatial-temporal features using a graph convolutional network. A categorical layer is introduced to account for the different influences of dynamic agents, while vehicle dynamics constraints ensure the feasibility of predicted trajectories. We developed a new heterogeneous and dense urban unsignalized intersection dataset (HID), capturing complex urban interactions, and conducted extensive experiments on HID, ApolloScape, and TRAF datasets. Results demonstrate that our model outperforms benchmark methods across diverse urban scenarios, and the integration of key modules significantly enhances prediction accuracy and performance.

PB-Trajectron: Physics bounded neural network for generalized trajectory prediction

Vehicle trajectory prediction is a critical technology in autonomous driving systems, as the quality of prediction directly affects downstream path planning and vehicle control. Although recent studies have shown that models combining kinematic rules with data-driven methods exhibit better interpretability in trajectory prediction, these models often adopt fixed constraint strengths, limiting their generalization ability in diverse traffic scenarios. This fixed constraint strength design restricts the model’s adaptability to different traffic environments.To address this issue, we propose PB-Trajectron, a dynamic integration mechanism that enhances the model’s prediction performance and generalization capability. PB-Trajectron is a dynamic integration mechanism that combines vehicle kinematic rules with deep learning prediction models for generalized vehicle trajectory prediction. The model integrates physics-based motion models and Deep Neural Networks (DNNs) to provide reasonable physical explanations for predicting vehicle trajectories at different speeds. First, we establish two vehicle kinematic constraints applicable to different prediction scenarios, enabling the agent to switch between these constraints based on the causal relationships between vehicle motion states to avoid generating non-physical trajectory results. Second, we propose a kinematic constraint decision framework based on velocity thresholds, which demonstrates adaptive adjustment, allowing the agent to switch tasks according to real-time state conditions and adaptively adjust the contribution of different physical constraints based on actual vehicle speeds. Finally, we investigate the advantages of PB-Trajectron in Out-of-Domain(OOD) generalization.Cross-scenario experiments on the nuScenes dataset show that, compared to previous methods, PB-Trajectron reduces the final displacement error by 7.69% when the prediction step length is 4. Furthermore, out-of-domain generalization tests on the INTERACTION dataset demonstrate that PB-Trajectron achieves a 12.23% reduction in average prediction error on datasets from different countries and scenarios compared to previous methods. The proposed mechanism can better adapt to complex and diverse traffic scenarios, laying the foundation for explainable and robust autonomous driving systems.

Multimodal vehicle trajectory prediction and integrated threat assessment algorithm based on adaptive driving intention

As autonomous driving and connected communication technologies advance swiftly, vehicle trajectory prediction has become increasingly significant. The motion of a vehicle is contingent not only on its historical trajectory but is also subject to the influence of surrounding vehicles, thereby exhibiting intricate social and temporal interdependencies. Furthermore, the inherent randomness and uncertainty in driver behavior render vehicle trajectory prediction inherently multimodal, a factor that is frequently neglected in current research. Against this backdrop, a multimodal vehicle trajectory prediction (MTP) model based on an encoder-decoder architecture is proposed to hierarchically extract historical features of vehicles. The model consists of five key components: temporal feature encoder module, spatial interaction module, spatial-temporal dependence module, driving intention fusion module and multimodal trajectory output module. Experiments on the NGSIM dataset show that the predictive performance of the model has been improved to varying degrees, especially at 3–5 s, where the improvement is more significant. Compared with state-of-the-art models, the Root Mean Square Error (RMSE) error at 5 s time horizon is 3.38 m on NGSIM dataset, which represents a 25 % improvement. To measure the safety of predicted trajectories, we propose a comprehensive threat assessment model that combines collision time (TTC), headway (TH) and time to lateral collision (TLC) metrics based on safe distance theory. This model not only evaluates the longitudinal collision threat in the following state, but also evaluates the lateral collision threat during driving maneuvers in multi lane scenarios, thereby comprehensively improving the safety of vehicle driving. This research also offers new perspectives and insights for the development of autonomous driving.

A POV-Based Highway Vehicle Trajectory Dataset and Prediction Architecture

A POV-Based Highway Vehicle Trajectory Dataset and Prediction Architecture

Counterfactual-based two-stage trajectory prediction framework for autonomous vehicle navigation systems

Abstract Accurate prediction of a vehicle’s trajectory is essential for autonomous driving and robotic operations in a complex, dynamic environment. However, current methods that rely on historical and environmental cues to predict future trajectories make the inner distance between testing and training environments often overlooked. To address this issue, we propose a counterfactual analysis method specifically designed for vehicle trajectory prediction. We begin by constructing a causal graph that includes historical trajectories, future trajectories, and environmental interactions. This graph captures the causal relationships between these elements. Our approach directly intervenes on the trajectory itself, effectively cutting off the inference from the environment to the trajectory. By doing so, we mitigate the negative impact of environmental bias. Furthermore, by comparing observed and intervened trajectory clues, we highlight the influence of environmental distance. Our goal is to improve the accuracy of trajectory prediction. Our counterfactual analysis module integrates seamlessly into various baseline prediction models. Experimental data demonstrates that our method outperforms others on publicly available Argoverse2 motion forecasting benchmarks.

MmSTCT: spatial–temporal convolution transformer network considering driving intention for multimodal vehicle trajectory prediction of highway

Due to the dynamic nature of traffic flow and the uncertainty of drivers' maneuvers, multimodal trajectory prediction remains challenging. In this paper, a multimodal vehicle trajectory prediction model is proposed based on the encoder-decoder structure, utilizing a spatial-temporal dilated causal convolutional transformer network for the hierarchical extraction of driving intention and spatial interaction features. Specifically, temporal information is extracted from the initial motion states by utilizing a bi-directional long short-term memory network. Subsequently, the social interaction module and spatial-temporal dependency module are introduced to model the interactive influences among surrounding vehicles and the target vehicle at the same timestamp as well as at different timestamps. Then adaptive fusion of driving intention is performed to generate multiple queries for accommodating the output of multimodal trajectory prediction module in a non-autoregressive manner. The effectiveness of the proposed model is demonstrated through experimental evaluations on the NGSIM, HighD and CitySim datasets.

Vehicle trajectory prediction based on cross-attention and multilevel spatio-temporal features

Accurately predicting vehicle trajectories is crucial for motion planning in autonomous driving. Some existing trajectory prediction models employ Long Short-Term Memory (LSTM) networks to encode trajectory information, but this approach may lead to information loss. Additionally, the sparsity of the grid representation may limit the model’s ability to extract vehicle interaction features. To address these issues, this paper proposes a vehicle trajectory prediction model based on cross-attention and multilevel spatio-temporal features. The proposed model incorporates a spatio-temporal feature extraction module designed to capture the dynamic evolution of vehicle states across both time and space. Furthermore, it employs a temporal cross-attention module that reuses hidden states to select spatio-temporal features closely related to the target vehicle’s historical state, thus compensating for the temporal information loss during the encoding process. Additionally, a spatial cross-attention module, informed by social context, is deployed to analyze the dynamic interactions among vehicles. This module filters spatio-temporal interaction features with a significant impact on the target vehicle’s trajectory. The model is trained and tested using the NGSIM dataset and the highD dataset. Compared with other models, the model presented in this paper shows higher prediction accuracy in long-term trajectory prediction.

Vehicle trajectory prediction method integrating spatiotemporal relationships with hybrid time-step scene interaction

In vehicle trajectory prediction, constructing the interactive relationships among vehicles within the traffic environment poses a significant challenge. Existing models predominantly focus on temporal dependencies within vehicle histories and spatial correlations among neighboring vehicles, overlooking the continuous influence of historical vehicle states on the current time step and the interplay of multiple sequences over time. To address these limitations, we propose a method for multimodal vehicle trajectory prediction that integrates Hybrid Time-step Scene Interaction (HTSI) into the spatiotemporal relationships. Firstly, we introduce the HTSI module, comprising Multi-step Temporal Information Aggregation (MTIA) and Single-step Temporal Information Aggregation (STIA) methods. MTIA utilizes multi-head attention mechanisms to capture temporal dependencies between consecutive frames, thereby generating new time series amalgamating the ongoing influence of historical time states on the current timestamp. Simultaneously, STIA employs multi-head attention mechanisms to capture the spatial dimension weights of multiple time series and, by aggregating spatial interaction features at each timestamp, generates new time series fused with spatial interaction influences. Subsequently, feature extraction is performed through LSTM layers. Moreover, we propose an improved DIPM pooling module, improving the model’s long-term prediction capability by selectively reusing historical hidden states. Ultimately, based on training results from the HighD and NGSIM datasets, our model demonstrates significant advantages in long-term prediction compared to other state-of-the-art trajectory prediction models. Specifically, within the 5 s prediction window, the model achieved a root mean square error (RMSE) of 2.79 m on the NGSIM dataset, representing a 33.62% improvement over the baseline model’s average accuracy. Additionally, on the HighD dataset, the model attained an RMSE of 2.16 m, reflecting a 33.43% enhancement. The crucial code can be obtained from the provided link: https://github.com/gyhhq/Prediction-trajectory .

Scenario-aware clustered federated learning for vehicle trajectory prediction with non-IID data

In recent years, Federated Learning (FL) has attracted much attention in Vehicle Trajectory Prediction (VTP) as it can resolve the critical issues of insufficient data, data privacy, and training efficiency. However, compared with centralized training, the model trained by FL may have insufficient prediction performance. This important issue comes from a statistical heterogeneity distribution of the local data in the participating clients, that is, non-IID. Therefore, this paper introduces a Clustered Federated Learning (CFL) approach for the VTP model to mitigate the influence of non-IID data. The proposed approach consists of federated trajectory clustering and federated VTP model training. In federated trajectory clustering, the optimal trajectory scenario discriminator is produced using federated K-means clustering without direct access to private data. In the federated VTP model training, multiple VTP models for specific trajectory scenarios are trained to deal with the influence of non-IID data. Experimental results reveal that our approach outperforms the state-of-the-art FL method on both NGSIM and HighD datasets, achieving up to 13.82% convergence acceleration and 12.47% RMSE reduction.

Vehicle Trajectory Prediction Based on Posterior Distributions Fitting and TCN-Transformer

Vehicle Trajectory Prediction Based on Posterior Distributions Fitting and TCN-Transformer

Two-Step Quadratic Programming for Physically Meaningful Smoothing of Longitudinal Vehicle Trajectories

Longitudinal vehicle trajectories suffer from errors and noise because of detection and extraction techniques, challenging their applications. Existing smoothing methods either lack physical meaning or cannot ensure solution existence and uniqueness. To address this, we propose a two-step quadratic programming method that aligns smoothed speeds and higher-order derivatives with physical laws, drivers’ behaviors, and vehicle characteristics. Unlike the well-known smoothing splines method, which minimizes a weighted sum of discrepancy and roughness in a single quadratic programming problem, our method incorporates prior knowledge of position errors into two sequential quadratic programming problems. Step 1 solves half-smoothed positions by minimizing the discrepancy between them and raw positions, subject to physically meaningful bounds on speeds and higher-order derivatives of half-smoothed positions. Step 2 solves smoothed positions by minimizing the roughness while maintaining physically meaningful bounds and allowing the deviations from raw data of smoothed positions by at most those of the half-smoothed positions and prior position errors. The second step’s coefficient matrix is not positive definite, necessitating the matching of the first few smoothed positions with corresponding half-smoothed ones, with equality constraints equaling the highest order of derivatives. We establish the solution existence and uniqueness for both problems, ensuring their well-defined nature. Numerical experiments using Next Generation Simulation (NGSIM) data demonstrate that a third-order derivative constraint yields an efficient method and produces smoothed trajectories comparable with manually re-extracted ones, consistent with the minimum jerk principle for human movements. Comparisons with an existing approach and application to the Highway Drone data set further validate our method’s efficacy. Notably, our method is a postprocessing smoothing technique based on trajectory data and is not intended for systematic errors. Future work will extend this method to lateral vehicle trajectories and trajectory prediction and planning for both human-driven and automated vehicles. This approach also holds potential for broader smoothing problems with known average error in raw data. History: This paper has been accepted for the Transportation Science Special Issue on ISTTT25 Conference. Funding: The authors extend their gratitude to the Pacific Southwest Region University Transportation Center and the University of California Institute of Transportation Studies (UC ITS) Statewide Transportation Research Program (STRP) for their valuable financial support.

Leveraging Cooperative Intent and Actuator Constraints for Safe Trajectory Planning of Autonomous Vehicles in Uncertain Traffic Scenarios

This study explores the integration of dynamic vehicle trajectories, vehicle safety factors, static traffic environments, and actuator constraints to improve cooperative intent modeling for autonomous vehicles (AVs) navigating uncertain traffic scenarios. Existing models often focus solely on interactions between dynamic trajectories, limiting their ability to fully interpret the intentions of surrounding vehicles. To address this limitation, we present a more comprehensive approach using the Cooperative Intent Multi-Layer Graph Neural Network (CMGNN) model. The CMGNN analyzes not only the dynamic trajectories but also the lane position relationships, vehicle angle changes, and actuator constraints and performs group interaction analysis. This richer information allows the CMGNN to more accurately capture the cooperative intent and better understand the surrounding vehicle behavior. This study investigated the impact of the CMGNN in the Carla simulator on surrounding vehicle trajectory prediction and AV safe trajectory planning. An innovative mechanism for dynamic trajectory risk assessment is introduced, which takes into account the constraints of the actuators when evaluating trajectory planning metrics. The results show that incorporating cooperative intent and considering the actuator limitations enhanced the CMGNN’s safety and driving efficiency in uncertain scenarios, significantly reducing the probability of AVs colliding. This is achieved as the model dynamically adapts its driving strategy based on the real-time traffic conditions, the perceived intentions of the surrounding vehicles, and the physical constraints of the vehicle actuators.

Car-following informed neural networks for real-time vehicle trajectory imputation and prediction

Vehicle trajectory information is a crucial part of improving the efficiency and the safety of the ITS. Data missing or irregular sampling in the real-world road traffic makes it hard to obtain accurate real-time vehicle trajectories. In this paper, we focus on trajectory imputation and prediction tasks with small data (magnitude set as 10 1 and 10 2 ). Limited by insufficient data, the simulation results of the existing data-driven algorithms are unsatisfactory. With car-following models integrated as prior physical information to constrain the training process, we design the car-following-informed neural network (CFINN). A multi-head self-attention layer is attached to the fully connected network layer to extract vehicle features. Different from the structure of most neural networks in regression analysis, an extra physics-based dataset is constructed in the CFINN. The loss function consists of two parts including the given trajectory's fitting error and the generated trajectory's residual error. We embed the gated recurrent unit-based encoder–decoder layer to the CFINN framework for trajectory predictions. The rationality and the superiority of our model are validated on the NGSIM dataset and the HighD dataset. Compared with baseline models in both single-vehicle and queue-typed trajectory imputation experiments, lower error can be achieved via the CFINN and coefficients of car-following models can be calibrated. According to driving regimes derived from CFINN-based trajectory prediction experiments, we discuss the impact of cut-in behaviours on the target vehicle and carry out the kinetic analysis. The novel neural network model driven by both data and physical knowledge provides technical support in vehicle status assessments and trajectory predictions.