UAV Sense and Avoid

Overview

As a response to continuous operations of the US Navy throughout the world, there is a growing need for autonomous improvement. One area of interest that is lacking modernization is current fleet logistics. This involves the movement of equipment and supplies from one ship or land point to another. As a result of this, a lightweight, cost effective, and transferable machine learning platform was designed using the NVIDIA Jetson Xavier with the goal of detecting and autonomously responding to obstacles. This will be mounted on a UAV platform.

Objectives

Within the scope of this project, the expectations are as follows:

Develop a visual based system that should be able to identify items in the environment and provide a response to detected objects

Integration using the NVIDIA Jetson Xavier system is required

Approach

The steps performed to complete this project can be found below:

1. Sponsor expectations were introduced and parameters of project were established.

2. Received all necessary components including the NVIDIA Jetson Xavier and camera system.

3. Initialized Jetson, flashed OS, and installed necessary languages and libraries.

4. Presented preliminary design to NAWCAD, approved alternative deliverables.

5. CAD models of several iterations of design were drawn and assembled.

6. Defined x-y dimensions for frames detected by sensor, represented position as binary and assigned object position to output via GPIO pins.

7. Programmed LED array to respond to object detection; represents UAV response.

8. Connected Jetson to Arduino and performed troubleshooting procedures to ensure a flawless connection.

9. Ensured self sufficient programming, needing no connection to host computer.

Outcomes

The LED matrix provides an alternative representation to the UAV response when detecting objects

Final design is lightweight and transferable; capable of being mounted on a UAV platform

The machine learning can be trained to detect different objects (i.e., birds, ships, etc.)