Angle Robustness Unmanned Aerial Vehicle Navigation in GNSS-Denied Scenarios
Authors: Yuxin Wang, Zunlei Feng, Haofei Zhang, Yang Gao, Jie Lei, Li Sun, Mingli Song
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiment results demonstrate that the proposed model outperforms the state-of-the-art by achieving improvements of 26.0% and 45.6% in the success rate of arrival under ideal and disturbed circumstances, respectively. To evaluate the vision-based navigation methods, we collect a new dataset termed as UAV AR368. Furthermore, we design the Simulation Flight Testing Instrument (SFTI) using Google Earth to simulate different flight environments, thereby reducing the expenses associated with real flight testing. |
| Researcher Affiliation | Academia | Yuxin Wang1, Zunlei Feng1, Haofei Zhang1, Yang Gao1, Jie Lei2, Li Sun3*, Mingli Song1 1Zhejiang University 2Zhejiang University of Technology 3Ningbo Innnovation Center, Zhejiang University |
| Pseudocode | No | The paper describes the architecture and modules but does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide an explicit statement about releasing its source code for the methodology or a link to a code repository. |
| Open Datasets | No | We collect a new dataset termed as UAV AR368 containing 368 routes and 56,880 images on the topic of UAV point-to-point navigation. To our knowledge, there are no publicly available datasets for UAV point-to-point navigation tasks. |
| Dataset Splits | No | Finally, the dataset is classified into 100 categories based on geographical positions and split into a training set with 45,398 images and a testing set with 11,482 images. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments. |
| Software Dependencies | No | The paper mentions 'imgaug (Jung et al. 2020) library' and 'Adam W optimizer' but does not specify version numbers for these or other key software components like programming languages or deep learning frameworks. |
| Experiment Setup | Yes | The value of K is 5 (excluding Xe). All input images are randomly cropped and resized to 224 224. The dimensions of RI k, RP k , Zk, Z k are 576, 512, 512, 512. The dimension, depth, and hidden dimension of the Cross-knowledge Attention-guided Module are 512, 4, and 1024. An eight-head strategy is adopted in the Masked Multi-Head Attention layer. The dropout rate is 0.1. The architecture is optimized by Adam W optimizer with a learning rate of 1e-3 and a weight decay of 0.1. The Cosine Annealing Warm Restarts is employed as the learning rate decay scheduler and Task-Uncertainty Loss (Kendall, Gal, and Cipolla 2017) is used as the loss function. The model is trained for 120 epochs with the batch size of 128. We conduct 100 and 160 flight tests with a length of over 5,800 m for ideal and disturbed environments. D is set to 30 m. |