GATraj: A Graph- and Attention-based Multi-Agent Trajectory Prediction Model

Cheng, Hui‐Ming; Liu, Mengmeng; Chen, Lin; Broszio, Hellward; Sester, Monika; Yang, Michael Ying

doi:10.48550/arxiv.2209.07857

Cited by 3 publications

(4 citation statements)

References 38 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The calculation of NC is based on trajectories with the best ADE. Following prior works [44], [9], [32], [11], we adopt the leave-one-out strategy for the evaluation.…”

Section: Evaluation Metrics and Protocolmentioning

confidence: 99%

ForceFormer: Exploring Social Force and Transformer for Pedestrian Trajectory Prediction

Zhang

Cheng

Johora

et al. 2023

2023 IEEE Intelligent Vehicles Symposium (IV)

View full text Add to dashboard Cite

Predicting trajectories of pedestrians based on goal information in highly interactive scenes is a crucial step toward Intelligent Transportation Systems and Autonomous Driving. The challenges of this task come from two key sources:(1) complex social interactions in high pedestrian density scenarios and ( 2) limited utilization of goal information to effectively associate with past motion information. To address these difficulties, we integrate social forces into a Transformerbased stochastic generative model backbone and propose a new goal-based trajectory predictor called ForceFormer. Differentiating from most prior works that simply use the destination position as an input feature, we leverage the driving force from the destination to efficiently simulate the guidance of a target on a pedestrian. Additionally, repulsive forces are used as another input feature to describe the avoidance action among neighboring pedestrians. Extensive experiments show that our proposed method achieves on-par performance measured by distance errors with the state-of-the-art models but evidently decreases collisions, especially in dense pedestrian scenarios on widely used pedestrian datasets.

show abstract

“…The calculation of NC is based on trajectories with the best ADE. Following prior works [44], [9], [32], [11], we adopt the leave-one-out strategy for the evaluation.…”

Section: Evaluation Metrics and Protocolmentioning

confidence: 99%

ForceFormer: Exploring Social Force and Transformer for Pedestrian Trajectory Prediction

Zhang

Cheng

Johora

et al. 2023

2023 IEEE Intelligent Vehicles Symposium (IV)

View full text Add to dashboard Cite

show abstract

“…Our model is trained to observe 8 time steps and predict the next 12-step trajectories with 20 modes. We adopt the standard leave-one-out training and test partitioning technique, wherein four out of the five subsets are utilized for training purposes, while the remaining subset is held out for testing [4]. The Argoverse dataset contains 323557 real-world driving scenarios and is split into training, validation, and test sets, with 205942, 39472, and 78143 samples.…”

Section: Datasetsmentioning

confidence: 99%

“…The "encoder-interactor-decoder" structure is the most popular architecture in trajectory prediction works of recent years [1][2][3][4]. Usually, the encoder aggregates the agent and con-text features from the input trajectory and high-definition (HD) map data, than input the encoding into the interactor to get the social attention and agent-lane attention, and finally output the prediction output via the decoder.…”

Section: Introductionmentioning

confidence: 99%

Q‐EANet: Implicit social modeling for trajectory prediction via experience‐anchored queries

Chen,

Wang,

Wang

et al. 2023

IET Intelligent Trans Sys

View full text Add to dashboard Cite

Accurately predicting the future trajectory and behavior of traffic participants is crucial for the maneuvers of self‐driving vehicles. Many existing works employed a learning‐based “encoder‐interactor‐decoder” structure, but they often fail to clearly articulate the relationship between module selections and real‐world interactions. As a result, these approaches tend to rely on a simplistic stacking of attention modules. To address this issue, a trajectory prediction network (Q‐EANet) is presented in this study, which integrates GRU encoders, MLPs and attention modules. By introducing a new explanatory rule, it makes a contribution to interpretable modeling, models the entire trajectory prediction process via an implicit social modeling formula. Inspired by the anchoring effect in decision psychology, the prediction task is formulated as an information query process that occurs before traffic participants make decisions. Specifically, Q‐EANet uses GRUs to encode features and utilizes attention modules to aggregates interaction information for generating the target trajectory anchors. Then, queries are introduced for further interaction. These queries, along with the trajectory anchors with added Gaussian noise, are then processed by a GRU‐based decoder. The final prediction results are obtained through a Laplace MDN. Experimental results on the several benchmarks demonstrate the effectiveness of Q‐EANet in trajectory prediction tasks. Compared to the existing works, the proposed method achieves state‐of‐the‐art performance with only simple module design. The code for this work is publicly available at https://github.com/Jctrp/socialea.

show abstract

“…Recent years, deep learning methods, such as Generative Adversarial Network (GAN) [23], Conditional Variational AutoEncoder (CVAE) [5,24], and attention mechanisms [25,26], have been introduced to the trajectory prediction task. Gupta et al [27] propose a fusion of LSTM and GAN, using the global pooling module of LSTM as the encoder-decoder generator and a discriminator composed of multiple LSTMs.…”

Section: B Trajectory Predictionmentioning

confidence: 99%

An end-to-end framework for pulmonary nodule detection and false positive reduction from CT Images

Chen

Cao

Dou

et al. 2020

The Fourth International Symposium on Image Computing and Digital Medicine

View full text Add to dashboard Cite

Perception that involves multi-object detection and tracking, and trajectory prediction are two major tasks of autonomous driving. However, they are currently mostly studied separately, which results in most trajectory prediction modules being developed based on ground truth trajectories without taking into account that trajectories extracted from the detection and tracking modules in real-world scenarios are noisy. These noisy trajectories can have a significant impact on the performance of the trajectory predictor and can lead to serious prediction errors. In this paper, we build an end-to-end framework for detection, tracking, and trajectory prediction called ODTP (Online Detection, Tracking and Prediction). It adopts the state-of-the-art online multi-object tracking model, QD-3DT, for perception and trains the trajectory predictor, DCENet++, directly based on the detection results without purely relying on ground truth trajectories. We evaluate the performance of ODTP on the widely used nuScenes dataset for autonomous driving. Extensive experiments show that ODPT achieves high performance end-to-end trajectory prediction. DCENet++, with the enhanced dynamic maps, predicts more accurate trajectories than its base model. It is also more robust when compared with other generative and deterministic trajectory prediction models trained on noisy detection results.

show abstract

GATraj: A Graph- and Attention-based Multi-Agent Trajectory Prediction Model

Cited by 3 publications

References 38 publications

ForceFormer: Exploring Social Force and Transformer for Pedestrian Trajectory Prediction

ForceFormer: Exploring Social Force and Transformer for Pedestrian Trajectory Prediction

Q‐EANet: Implicit social modeling for trajectory prediction via experience‐anchored queries

An end-to-end framework for pulmonary nodule detection and false positive reduction from CT Images

Contact Info

Product

Resources

About