Trajectory Prediction Model Research Articles

CCTR: Calibrating Trajectory Prediction for Uncertainty-Aware Motion Planning in Autonomous Driving

Autonomous driving systems rely on precise trajectory prediction for safe and efficient motion planning. Despite considerable efforts to enhance prediction accuracy, inherent uncertainties persist due to data noise and incomplete observations. Many strategies entail formalizing prediction outcomes into distributions and utilizing variance to represent uncertainty. However, our experimental investigation reveals that existing trajectory prediction models yield unreliable uncertainty estimates, necessitating additional customized calibration processes. On the other hand, directly applying current calibration techniques to prediction outputs may yield sub-optimal results due to using a universal scaler for all predictions and neglecting informative data cues. In this paper, we propose Customized Calibration Temperature with Regularizer (CCTR), a generic framework that calibrates the output distribution. Specifically, CCTR 1) employs a calibration-based regularizer to align output variance with the discrepancy between prediction and ground truth and 2) generates a tailor-made temperature scaler for each prediction using a post-processing network guided by context and historical information. Extensive evaluation involving multiple prediction and planning methods demonstrates the superiority of CCTR over existing calibration algorithms and uncertainty-aware methods, with significant improvements of 11%-22% in calibration quality and 17%-46% in motion planning.

An Adaptive Multimodal Data Vessel Trajectory Prediction Model Based on a Satellite Automatic Identification System and Environmental Data

Ship trajectory prediction is essential for ensuring safe route planning and to have advanced warning of the dangers at sea. With the development of deep learning, most of the current research has explored advanced prediction methods based on historical spatio-temporal Automatic Identification System (AIS) data. However, environmental factors such as sea wind and visibility also affect ship navigation in real-world maritime shipping. Therefore, developing reliable models utilizing multimodal data, such as AIS and environmental data, is challenging. In this research, we design an adaptive multimodal vessel trajectory data prediction model (termed AMD) based on satellite AIS and environmental data. The AMD model mainly consists of an AIS-based extraction network, an environmental-based extraction network, and a fusion block. In particular, this work considers multimodal data such as historical spatio-temporal information and environmental factors. Time stamps and distances are correlated with AIS and environmental data, and a multilayer perceptron and gated recurrent unit networks are used to design multimodal feature extraction networks. Finally, the fusion block realizes the fusion output of multimodal features to improve the reliability of the AMD model. Several quantitative and qualitative experiments are conducted using real-world AIS and multimodal environmental datasets. Numerous experimental results prove that prediction performance using multimodal data can ensure satisfactory accuracy and reliability while exhibiting a positive impact on improving maritime transport services.

Enhancing lane changing trajectory prediction on highways: A heuristic attention-based encoder-decoder model

Accurate prediction of lane changing (LC) trajectories plays a vital role in ensuring safe and efficient traffic flow on highways. This paper proposes a LC trajectory prediction model based on encoder-decoder architecture to address low long-term prediction accuracy problem and to gain insight into the underlying motivations of LC behavior. Three specific enhancements were proposed to improve the performance of encoder-decoder. The first enhancement involves the utilization of Neighborhood Component Analysis (NCA) to identify the most relevant input features from driving environment. The second enhancement involves the introduction of heuristic attention to capture intricate time-dependent and environment-dependent patterns of input data. Finally, a trajectory check module was proposed to supervise and adjust the prediction outputs based on vehicle dynamics constraints. The proposed model was trained and evaluated using the HighD dataset. Experimental results demonstrated that the proposed model outperforms improved encoder-decoder architectures designed by previous studies in terms of prediction accuracy in long-term prediction, which the lateral and longitude MAE were 0.041 m and 1.842 m in 5 s prediction duration, respectively. Additionally, it was observed that preceding vehicles in the current and target lane, and following vehicle in the target lane, exert a significantly influential effect on LC maneuvers, which represents the underlying motivations behind LC behavior. The findings of this study contribute to the advancement of intelligent transportation systems (ITS) and autonomous vehicles (AVs) by providing an effective LC trajectory prediction.

Real-Time UAV Trajectory Prediction for UTM Surveillance Using Machine Learning

The Unmanned Aerial System (UAS) Traffic Management (UTM) surveillance plays an important role in monitoring the safety compliance of UAV flights within specific operational zones. However, the quality of data reception from UAVs depends on transmission signal quality, leading to variable data latency during flights. As a means of addressing this issue and improving UAV operational safety, trajectory prediction emerges as a promising solution. This study aims to implement real-time trajectory prediction through the integration of machine learning techniques within the UTM surveillance system using Broadcast Remote ID. To facilitate this, a homemade receiver for broadcast Remote ID is developed using the ESP32 microcontroller. Subsequently, two distinctive flight tests are executed using a DJI Phantom 4 UAV. In the first flight, UAV trajectory data are used to serve as training input for the machine learning algorithms. The second flight focuses on the implementation of real-time trajectory prediction. The trajectory prediction model uses inputs such as latitude, longitude, height, speed, direction, latency time, and the Received Signal Strength Indicator (RSSI). Through an analysis of the first flight’s offline data, the Gated Recurrent Unit (GRU) algorithm is preferred to the Long Short-Term Memory (LSTM) algorithm. The outcomes of the second flight test prove the feasibility of the GRU-based trajectory prediction. The GRU model successfully produces real-time predictions during UAV flight, showcasing accuracy levels similar to those derived from offline prediction analyses with short processing time. This real-time trajectory prediction could improve the safety of UAV operation by providing the estimated position when the latency time is higher than the threshold.

Predicting vehicle trajectory of non-lane based driving behaviour with Temporal Fusion Transformer

Complex road sections without lane markings cannot constrain vehicles to follow the lane disciplines. As a result, vehicles often exhibit more disorderly rapid lateral movements (RLMs) in these areas, making it difficult to accurately predict vehicle trajectories. This study takes toll plaza diverging area as an example to propose a framework incorporated the Hidden Markov Model (HMM) and Temporal Fusion Transformer (TFT) for vehicle trajectory prediction in non-lane based complex road sections. The results demonstrate that the vehicles exhibit more RLMs when there are more toll lanes matching their toll collection types. Validated on two toll plaza diverging areas with different structures, the proposed framework achieves higher prediction accuracy than other state-of-the-art predictive methods, particularly in long prediction horizons. In addition, the interpretability of TFT suggests that incorporating RLM intention prediction and environmental factors specific to non-lane based areas into trajectory prediction is of great importance.

Ship collision avoidance decision-making research in coastal waters considering uncertainty of target ships

Ship collision avoidance has always been a concern and it is crucial for achieving safe navigation of ships at sea. There are many studies on ship collision avoidance in open water, but less attention on coastal waters considering the uncertainty of target ships due to the complexity of the environment and traffic flow. In this paper, collision avoidance decision-making research in coastal waters considering the uncertainty of target ships was proposed. Firstly, accurate ship trajectories are obtained by preprocessing the raw Automatic Identification System (AIS) data. Subsequently, the processed trajectories are clustered using the Ordering Points to Identify the Clustering Structure (OPTICS) algorithm and Hausdorff distance, acquiring a dataset for trajectory prediction of target ships. Then, a mixed Gaussian model is utilized to calculate the prior probability distribution of the prediction model, thus establishing a trajectory prediction model that considers the uncertainty of the target ship. Finally, ship maneuverability is simulated using the Mathematical Model Group (MMG) and Proportion Integration Differentiation (PID) models, and a collision avoidance decision-making model for ships is constructed. The proposed algorithm has been tested and verified in a case study. The results show that the approach effectively predicts the trajectory of the target ship and facilitates informed collision avoidance decision-making.

The attentional guidance and facilitating effects of group behavioral cues on individual college pedestrians’ jaywalking decisions

Objective Jaywalking is an important cause of pedestrian-related automobile accidents. Exploring the factors that influence jaywalking behavior and suggesting appropriate improvement measures are critical for reducing automobile accidents involving pedestrians. Methods This study divided traffic situations into high-risk and low-risk situations. Each situation contained three visual attention cues: vehicle, traffic light, and group behavior. Based on this, the role of visual cues in guiding pedestrians’ attention and influencing their decisions during jaywalking was examined. Sixty participants, with an average age of 19, were recruited. They were shown 84 crosswalk videos randomly while their crossing decisions and eye movement data were recorded. Results In low-risk situations, pedestrians spent more attention on group behavioral cues when making jaywalking decisions. The rate of jaywalking increased with the number of other jaywalking pedestrians. In high-risk situations, the pedestrians’ total fixation duration at vehicle hazard cues was longer when making jaywalking decisions, and the jaywalking rate decreased. Conclusions The results indicate that pedestrians’ jaywalking decisions were based on other pedestrians’ illegal crossing cues and automatic associative processes in low-risk situations. The higher the number of people crossing the street, the higher the number of pedestrians illegally crossing the road. In high-risk situations, pedestrians paid more attention to vehicle hazard cues before making jaywalking decisions, and fewer illegal crossings. The jaywalking decisions were based on a risk assessment, a controlled analytical process. The results verify the effect of visual cues on pedestrians’ attentional guidance and decision-making in different traffic situations, as well as the effectiveness of visual attention in predicting decision intention. The findings provide a theoretical basis and data reference for pedestrian safety education and constructing an intelligent driving pedestrian trajectory prediction model.

Neural differential constraint-based pedestrian trajectory prediction model in ego-centric perspective

Autonomous vehicles offer significant advantages for transportation systems, particularly in enhancing traffic safety. To achieve this goal, it is crucial to comprehensively understand and predict the future trajectories of pedestrians in proximity to autonomous vehicles. Many contemporary approaches for predicting pedestrian trajectories heavily rely on neural networks, especially recurrent neural networks. However, these approaches do not explicitly incorporate the dynamics of pedestrian movement and instead rely on data-driven black-box models. Consequently, these models may fall short in terms of interpretability and fail to adhere to the fundamental principles of kinematics. In response to these limitations, our work introduces an innovative model for pedestrian trajectory prediction grounded in neural differential constraints. We aim to investigate temporal changes in pedestrian state variables, such as position and speed, using neural networks. During the prediction process, the output of the neural network is governed by differential equations. This approach ensures that the generated trajectories align with the fundamental principles of physics, harnessing the combined power of neural networks and physics-based pedestrian motion models. Furthermore, our research endeavors to develop a cohesive framework that seamlessly integrates pedestrian movement patterns with the influence of ego-vehicles, while also considering potential destinations to inform future trajectory planning. We conducted extensive experiments on two publicly available real-world datasets to assess the effectiveness of our model in enhancing prediction accuracy and providing coherent explanations of pedestrian motion, comparing it to state-of-the-art methods.

Multi-modality trajectory prediction with the dynamic spatial interaction among vehicles under connected vehicle environment

Compared to non-connected vehicle environments, the connected vehicle environment establishes vehicle interconnection through communication technologies, resulting in more complex interaction, network topologies, and large-scale inputs. This complexity renders traditional trajectory prediction models, which rely primarily on inputting historical information of the target vehicle, inadequate for handling the complex and dynamic interactive lane-changing scenarios in connected vehicle environments. In a connected vehicle environment, it is necessary to propose a more targeted and stable lane-changing behavior prediction method based on vehicle traveling characteristics. Taking into account dynamic spatial interaction among vehicles, this study proposes a multi-modality trajectory prediction model called STA-LSTM to perform analysis on the potential interactive behaviors among vehicles under connected vehicle lane-changing scenarios, and specifically to expand the multi-modality feature input of the vehicle trajectory prediction model. The spatial grid occupancy method is used to model the interactions between vehicles. A space-dimensional attention mechanism is introduced to adaptively match the influencing weights of the surrounding vehicles with the target vehicle and to improve the interactive information extraction method. In addition, the attention module is incorporated into the LSTM decoder from the time dimension so that the established model can identify significant historical hidden features during each trajectory decoding process. To account for the uncertainty of trajectory prediction, the vectors of vehicle interactions are incorporated into contextual information to improve the reliability of prediction results and the robustness of the established model. Compared with conventional baseline models, the proposed model exhibited lower root mean square error (RMSE) and negative log-likelihood (NLL) values, and the RMSE values at different prediction times of 1s, 2s, 3s, 4s, and 5s are 0.46m, 1.15m, 1.89m, 2.84m, and 4.05m, respectively. This indicates that the proposed model can accurately predict the interactions between vehicles and the travel paths of surrounding target vehicles.

A Pedestrian Trajectory Prediction Method for Generative Adversarial Networks Based on Scene Constraints

Pedestrian trajectory prediction is one of the most important topics to be researched for unmanned driving and intelligent mobile robots to perform perceptual interaction with the environment. To solve the problem of the SGAN (social generative adversarial networks) model lacking an understanding of pedestrian interaction and scene constraints, this paper proposes a trajectory prediction method based on a scenario-constrained generative adversarial network. Firstly, a self-attention mechanism is added, which can integrate information at every moment. Secondly, mutual information is introduced to enhance the influence of latent code on the predicted trajectory. Finally, a new social pool is introduced into the original trajectory prediction model, and a scene edge extraction module is added to ensure the final output path of the model is within the passable area in line with the physical scene, which greatly improves the accuracy of trajectory prediction. Based on the CARLA (CAR Learning to Act) simulation platform, the improved model was tested on the public dataset and the self-built dataset. The experimental results showed that the average moving deviation was reduced by 26.4% and the final offset was reduced by 23.8%, which proved that the improved model could better solve the uncertainty of pedestrian turning decisions. The accuracy and stability of pedestrian trajectory prediction are improved while maintaining multiple modes.

Motion Planning for Autonomous Driving with Real Traffic Data Validation

Accurate trajectory prediction of surrounding road users is the fundamental input for motion planning, which enables safe autonomous driving on public roads. In this paper, a safe motion planning approach is proposed based on the deep learning-based trajectory prediction method. To begin with, a trajectory prediction model is established based on the graph neural network (GNN) that is trained utilizing the INTERACTION dataset. Then, the validated trajectory prediction model is used to predict the future trajectories of surrounding road users, including pedestrians and vehicles. In addition, a GNN prediction model-enabled motion planner is developed based on the model predictive control technique. Furthermore, two driving scenarios are extracted from the INTERACTION dataset to validate and evaluate the effectiveness of the proposed motion planning approach, i.e., merging and roundabout scenarios. The results demonstrate that the proposed method can lower the risk and improve driving safety compared with the baseline method.

Multi domain decomposed reduced order model for store trajectory prediction

Multi domain decomposed reduced order model for store trajectory prediction

A Multi-Task Learning Model for Trajectory and Cellular Signal Power Prediction

A Multi-Task Learning Model for Trajectory and Cellular Signal Power Prediction

An Empirical-informed Model for the Early Degradation Trajectory Prediction of Lithium-ion Battery

An Empirical-informed Model for the Early Degradation Trajectory Prediction of Lithium-ion Battery

Vehicle Lane Change Multistep Trajectory Prediction Based on Data and CNN_BiLSTM Model

In order to accurately predict the lane‐changing trajectory of the vehicle and improve the driving safety of the vehicle, a lane‐changing trajectory prediction model based on the combination of convolutional neural network (CNN) and bidirectional long short‐term memory (BiLSTM) neural network is proposed by comprehensively considering the historical driving behavior, the spatial characteristics of surrounding vehicles and the bidirectional time sequence information of the vehicle trajectory. Firstly, the vehicle trajectory data are filtered and smoothed, and it is divided into three categories: left lane change, right lane change, and straight driving, and a lane change trajectory sample set is constructed. Secondly, CNN‐BiLSTM model is constructed to identify the sample set of lane‐changing trajectory. Considering the interaction between vehicles in the driving process, the information of predicted vehicle, and surrounding vehicles is taken as the input of the model. The extracted feature vector is input to the BiLSTM layer for prediction after the CNN layer feature extraction, and the horizontal and vertical coordinates of the target vehicle at the next time are output. Thirdly, the trajectory data of the US‐101 dataset in NGSIM is selected to verify the performance of the CNN‐BiLSTM model, and at the same time, it is compared with models such as CNN‐LSTM, long short‐term memory (LSTM), BiLSTM, and CNN‐GRU‐Att. Finally, the verification result shows that the overall fitting degree of the vehicle lane change trajectory prediction of the proposed model reaches 99.50%, and the mean square error and mean absolute error are 0.0003076 and 0.01417, which are improved compared with other models. In the meanwhile, the research on multistep trajectory prediction in different prediction time domains is carried out. It was found that the longer the prediction time domain is, the lower the prediction performance of the model decreases, but the prediction accuracy still reached more than 96%, and it was able to accurately predict the lane change trajectory.

SFC active reconfiguration based on user mobility and resource demand prediction in dynamic IoT-MEC networks.

To achieve secure, reliable, and scalable traffic delivery, request streams in mobile Internet of Things (IoT) networks supporting Multi-access Edge Computing (MEC) typically need to pass through a service function chain (SFC) consisting of an ordered series of Virtual Network Functions (VNFs), and then arrive at the target application in the MEC for processing. The high mobility of users and the real-time variability of network traffic in IoT-MEC networks lead to constant changes in the network state, which results in a mismatch between the performance requirements of the currently deployed SFCs and the allocated resources. Meanwhile, there are usually multiple instances of the same VNF in the network, and proactively reconfiguring the deployed SFCs based on the network state changes to ensure high quality of service in the network is a great challenge. In this paper, we study the SFC Reconfiguration Strategy (SFC-RS) based on user mobility and resource demand prediction in IoT MEC networks, aiming to minimize the end-to-end delay and reconfiguration cost of SFCs. First, we model SFC-RS as Integer Linear Programming (ILP). Then, a user trajectory prediction model based on codec movement with attention mechanism and a VNF resource demand prediction model based on the Long Short-Term Memory (LSTM) network are designed to accurately predict user trajectories and node computational and storage resources, respectively. Based on the prediction results, a Prediction-based SFV Active Reconfiguration (PSAR) algorithm is proposed to achieve seamless SFC migration and routing update before the user experience quality degrades, ensuring network consistency and high quality service. Simulation results show that PSAR provides 51.28%, 28.60%, 21.75%, and 16.80% performance improvement over the existing TSRFCM, DDQ, OSA, and DPSM algorithms in terms of end-to-end delay reduction, and 33.32%, 18.94%, 67.42%, and 60.61% performance optimization in terms of reconfiguration cost reduction.

An Enhanced Vehicle Trajectory Prediction Model Leveraging LSTM and Social-Attention Mechanisms

An Enhanced Vehicle Trajectory Prediction Model Leveraging LSTM and Social-Attention Mechanisms

When Is It Likely to Fail? Performance Monitor for Black-Box Trajectory Prediction Model

When Is It Likely to Fail? Performance Monitor for Black-Box Trajectory Prediction Model

IMA‐LSTM: An Interaction‐Based Model Combining Multihead Attention with LSTM for Trajectory Prediction in Multivehicle Interaction Scenario

The rapid development of vehicle‐to‐vehicle (V2V) communication technology provides more opportunities to improve traffic safety and efficiency, which facilitates the exchange of multivehicle information to mine potential patterns and hidden associations in vehicle trajectory prediction. To address the importance of fine‐grained vehicle interaction modelling in vehicle trajectory prediction, the present work proposes an integrated vehicle trajectory prediction model that combines the multihead attention mechanism with long short‐term memory (IMA‐LSTM) in multivehicle interaction scenarios. Compared to existing studies, a dedicated feature extraction module including both individual features and interactive features is designed and sophisticated multihead attention mechanism is applied with LSTM framework to capture the variation of spatial‐temporal interactions between vehicles. The performance of the proposed model in different scenarios is examined using both the high‐D and the NGSIM dataset through comprehensive comparison experiments. The results indicate that the proposed IMA‐LSTM model presents great improvement in vehicle trajectory prediction performance in different scenarios compared to the model that does not consider multivehicle interaction features. Moreover, it outperforms other existing models in 3–5s prediction horizons and such superiority is more evident in left lane‐changing (LLC) scenarios than lane‐keeping (LK) and right lane‐changing (RLC) scenarios. The outcomes fully address the importance of fine‐grained interactive feature modelling in improving vehicle trajectory performance in complex multivehicle interaction scenarios and could further contribute to more refined traffic safety and traffic efficiency management.

Kinetic analysis of efficient energy conversion in table tennis sports

Abstract Improving the accuracy and stability of players’ stroke speed and stroke landing point and maximizing the energy conversion have become important tasks in table tennis skill practice. This paper analyzes the forces during ball collision and the energy conversion law of its motion state and proposes to measure and identify the key mechanical parameters of table tennis motion from the aspects of batting speed, rotation, and batting angle. The 3D coordinates of the ball are reconstructed, and the parameters are measured in the translational and rotational states of the ball, respectively. The UKF state estimator is used to construct the system process equations of the table tennis ball motion model, and the mathematical expressions are further derived to obtain the kinematic-based post-collision trajectory prediction model of the table tennis ball. The accuracy of the model in predicting the trajectory of the table tennis ball movement is evaluated by the distribution of the error points of the collision points of the table tennis ball, and it is known through experiments that the range of the error of the collision point prediction of the table tennis ball by the method used in this paper is in the range of (x±0.05, y±0.05) m, which is smaller than that estimated by the traditional table tennis ball physical model, and the model’s prediction accuracy is good. After analyzing the motion trajectories with different angular velocities, drag coefficients, initial velocities, and exit angles, the optimal hitting speed of a curved circle ball is obtained as V=15m/s.