Abstract
The increased demand for Unmanned Aerial Vehicles (UAV) has also led to higher demand for realistic and efficient UAV testing environments. The current use of simulated environments has been shown to be a relatively inexpensive, safe, and repeatable way to evaluate UAVs before real-world use. However, the use of generic environments and manually-created custom scenarios leaves more to be desired. In this paper, we propose a new testbed that utilizes machine learning algorithms to procedurally generate, scale, and place 3D models to create a realistic environment. These environments are additionally based on satellite images, thus providing users with a more robust example of real-world UAV deployment. Although certain graphical improvements could be made, this paper serves as a proof of concept for an novel autonomous and relatively-large scale environment generator. Such a testbed could allow for preliminary operational planning and testing worldwide, without the need for on-site evaluation or data collection in the future.
Original language | English |
---|---|
Article number | 2185 |
Number of pages | 18 |
Journal | Applied Sciences |
Volume | 11 |
Issue number | 5 |
DOIs | |
Publication status | Published - 2 Mar 2021 |
Keywords
- Artificial intelligence
- Autonomous vehicles
- Deep learning
- Machine learning
- Neural network
- Real-world testbed
- Satellite images
- UAV