Article
Computer Science, Interdisciplinary Applications
Zhenning Li, Zhiwei Chen, Yunjian Li, Chengzhong Xu
Summary: This study presents a new multimodal trajectory prediction framework based on the transformer network to address the challenging task of predicting surrounding agent trajectories in heterogeneous traffic environments. The proposed framework includes a hierarchical-structured context-aware module and an efficient linear global attention mechanism. It also introduces a novel auxiliary loss to penalize infeasible off-road predictions. The empirical results demonstrate the state-of-the-art performance of the proposed model on the Lyft l5kit data set, enhancing the accuracy and feasibility of prediction outcomes.
COMPUTER-AIDED CIVIL AND INFRASTRUCTURE ENGINEERING
(2023)
Article
Engineering, Civil
Kunpeng Zhang, Xiaoliang Feng, Lan Wu, Zhengbing He
Summary: This study proposes a Graph Attention Transformer (Gatformer) method for predicting the future trajectories of surrounding traffic agents for autonomous vehicles. By utilizing Convolutional Neural Networks (CNNs) to extract spatial features and a position encoder to encode spatial and temporal features, and modeling the interactions using the Graph Attention Network (GAT) block, trajectories for multiple agents can be simultaneously predicted using a Transformer network.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2022)
Article
Engineering, Civil
Neel P. Bhatt, Amir Khajepour, Ehsan Hashemi
Summary: Predicting object motion behavior is a challenging and crucial task for safe decision making and path planning in autonomous vehicles. We propose MPC-PF, a trajectory predictor based on potential field, which incorporates social interaction and space considerations to address the limitations of existing models in terms of bias and accuracy across the prediction horizon.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Yusheng Peng, Gaofeng Zhang, Jun Shi, Benzhu Xu, Liping Zheng
Summary: Pedestrian trajectory prediction is an important research topic in computer vision, and this paper proposes an LSTM model based on social relation attention and interaction awareness to simulate social behavior during pedestrian walking. By using social relation features and attention mechanism, more accurate trajectory prediction is achieved.
Article
Engineering, Mechanical
Guoying Chen, Zheng Gao, Min Hua, Bin Shuai, Zhenhai Gao
Summary: This paper proposes a fusion algorithm that considers driving style and vehicle dynamics to improve the accuracy of lane change trajectory prediction for autonomous vehicles. The algorithm involves a long short-term memory lane change behavior recognition model and a Gaussian process motion modeling trajectory prediction algorithm.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2024)
Article
Computer Science, Information Systems
Salar Arbabi, Davide Tavernini, Saber Fallah, Richard Bowden
Summary: This paper introduces a planning framework for autonomous driving that learns multiple action policies from human-human interactions. The framework uses encoder-decoder recurrent neural networks and mixture density networks to model interactions and probability distributions over driver actions. It generates context-dependent candidate plans and predicts probable future plans of human drivers. The approach leverages fast computation on a graphic processing unit and is tested in a simulated highway driving environment.
Review
Engineering, Civil
Sajjad Mozaffari, Omar Y. Al-Jarrah, Mehrdad Dianati, Paul Jennings, Alexandros Mouzakitis
Summary: This article provides a comprehensive review of deep learning-based approaches for vehicle behavior prediction. It discusses the challenges and issues in behavior prediction and categorizes and reviews the most recent solutions based on input representation, output type, and prediction method. The article also evaluates the performance of several solutions and outlines potential future research directions.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2022)
Article
Computer Science, Information Systems
Luyao Ye, Zezhong Wang, Xinhong Chen, Jianping Wang, Kui Wu, Kejie Lu
Summary: Modeling interactions among vehicles is critical for improving efficiency and safety in autonomous driving. Most existing works consider interaction information implicitly and do not explore shared interaction representations. This article proposes a general graph self-attention network to learn interaction representations and utilizes pretraining and fine-tuning steps.
IEEE INTERNET OF THINGS JOURNAL
(2022)
Article
Engineering, Civil
Zihao Sheng, Yunwen Xu, Shibei Xue, Dewei Li
Summary: This paper proposes a graph-based spatial-temporal convolutional network (GSTCN) to predict future trajectory distributions of neighbor vehicles. The network combines graph convolutional network (GCN) and convolutional neural network (CNN) to capture spatial interactions and temporal features between vehicles, resulting in accurate trajectory predictions.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2022)
Article
Computer Science, Artificial Intelligence
Tao Huang, Rui Fu
Summary: A novel DFoA prediction method based on feature visualization of a deep autonomous driving model is proposed in this study, which does not require ground-truth DFoA data for training. By employing a multimodal spatiotemporal convolutional network and attention mechanism, the method accurately predicts DFoA by fusing features and utilizing ConvLSTM network for successive frames.
KNOWLEDGE-BASED SYSTEMS
(2022)
Article
Robotics
Li Wang, Tao Wu, Hao Fu, Liang Xiao, Zhiyu Wang, Bin Dai
Summary: Trajectory prediction is crucial for autonomous driving and mobile robots, with challenges in modeling actor-actor and actor-scene interactions, as well as considering different motion characteristics of each actor. The proposed method integrates multiple contextual cues, including motion features, interaction behaviors, and scene features, achieving effective results on two widely-used datasets.
IEEE ROBOTICS AND AUTOMATION LETTERS
(2021)
Article
Computer Science, Information Systems
Denggui Wang, Weiping Fu, Jincao Zhou, Qingyuan Song
Summary: Motion planning framework is proposed to enable autonomous vehicles to navigate safely through urban roads with occlusions, considering safety, comfort, and efficiency. The solution consists of local path planning, trajectory planning, and speed planning. The proposed model enhances autonomous vehicle comfort levels by about 32% $\sim $ 48% compared to the baseline method utilizing Automatic Emergency Braking system (AEB) in high pedestrian traffic scenarios. Simulation verification shows that the proposed model ensures safe autonomous driving in traffic scenarios with occlusions while maintaining comfort and efficiency.
Article
Engineering, Civil
Wenbo Shao, Yanchao Xu, Jun Li, Chen Lv, Weida Wang, Hong Wang
Summary: This study aims to address the uncertainty and lack of explainability in autonomous driving by exploring the impact of traffic environment on prediction algorithms. The study proposes a trajectory prediction framework with epistemic uncertainty estimation ability to output high uncertainty when facing unforeseeable or unknown scenarios. The results indicate that deep ensemble-based methods have advantages in improving robustness while estimating epistemic uncertainty.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2023)
Article
Engineering, Civil
Mengyin Fu, Ting Zhang, Wenjie Song, Yi Yang, Meiling Wang
Summary: This article analyzes NGSIM data and develops an LSTM-based framework to efficiently predict future trajectories of surrounding vehicles, projecting them into a spatio-temporal domain to create an octree map, resolving the issue of dynamic disturbances in autonomous driving.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2022)
Review
Green & Sustainable Science & Technology
Renbo Huang, Guirong Zhuo, Lu Xiong, Shouyi Lu, Wei Tian
Summary: This paper reviews recent works based on improvement designs and summarizes them based on three criteria: scene input representation, context refinement, and prediction rationality improvement. It discusses new occupancy flow prediction methods in addition to trajectory prediction, and outlines commonly used datasets, evaluation metrics, and potential research directions.