Human Observation-Inspired Trajectory Prediction for Autonomous Driving in Mixed-Autonomy Traffic Environments
This repository contains the official implementation of Human Observation-Inspired Trajectory Prediction for Autonomous Driving in Mixed-Autonomy Traffic Environments, accepted by the Association for the Advancement of Artificial Intelligence (ICRA) 2024.
- introduce a sophisticated pooling mechanism that replicates human attention allocation with a novel adaptive visual sector.
- Introduce a novel dynamic traffic graph to extract the interaction of agents using a unique topology graph structure constructed using Convolutional Neural Networks (CNN) and Graph Attention Networks (GAT).
- In benchmark tests on the NGSIM, HighD, and MoCAD datasets, our model outperforms SOTA baselines by at least 15.2%, 19.4% and 12.0%, respectively, demonstrating its impressive accuracy and applicability in various traffic scenarios, including highways and dense urban areas.
Human Observation-Inspired Trajectory Prediction for Autonomous Driving in Mixed-Autonomy Traffic Environments, accepted by the journal IEEE International Conference on Robotics and Automation (ICRA). (Camera-ready)
@article{liao2024human,
title={Human Observation-Inspired Trajectory Prediction for Autonomous Driving in Mixed-Autonomy Traffic Environments},
author={Liao, Haicheng and Liu, Shangqian and Li, Yongkang and Li, Zhenning and Wang, Chengyue and Wang, Bonan and Guan, Yanchen and Xu, Chengzhong},
journal={arXiv preprint arXiv:2402.04318},
year={2024}
}