BAT: Behavior-Aware Human-Like Trajectory Prediction for Autonomous Driving

Authors: Haicheng Liao, Zhenning Li, Huanming Shen, Wenxuan Zeng, Dongping Liao, Guofa Li, Chengzhong Xu

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate BAT s performance across the Next Generation Simulation (NGSIM), Highway Drone (High D), Roundabout Drone (Roun D), and Macao Connected Autonomous Driving (Mo CAD) datasets, showcasing its superiority over prevailing state-of-the-art (SOTA) benchmarks in terms of prediction accuracy and efficiency. Remarkably, even when trained on reduced portions of the training data (25%), our model outperforms most of the baselines, demonstrating its robustness and efficiency in predicting vehicle trajectories and the potential to reduce the amount of data required to train autonomous vehicles, especially in corner cases.
Researcher Affiliation Academia 1 University of Macau 2 University of Electronic Science and Technology of China 3 Peking University 4 Chongqing University 5 Tsinghua University
Pseudocode No The paper describes its model architecture and components but does not provide pseudocode or an algorithm block.
Open Source Code Yes The project page is available on our Git Hub. 1https://github.com/Petrichor625/BATraj-Behavior-aware Model
Open Datasets Yes We evaluate the effectiveness of our model using four datasets: NGSIM (Deo and Trivedi 2018), High D (Krajewski et al. 2018), Roun D (Krajewski et al. 2020), and Mo CAD. ... Mo CAD, set to be publicly available, is pivotal for research in right-hand-drive dynamics and enhancing trajectory prediction models.
Dataset Splits Yes These data sets were partitioned into training, validation, and test sets using standard sampling.
Hardware Specification Yes Our model is trained to converge using an NVIDIA A100 40GB GPU.
Software Dependencies No The paper does not provide specific version numbers for software dependencies or libraries.
Experiment Setup Yes The trajectories for the NGSIM, High D, and Mo CAD datasets were divided into 8-second intervals. The first 3 seconds served as the trajectory history (th = 3) for input, and the following 5 seconds represented the ground truth (tf = 5) for output. For the Roun D dataset, the trajectories were divided into 6-second chunks with th = 2 and tf = 4. ... We introduce the Negative Log-Likelihood criterion as a complement to the RMSE in the loss function.