Devon Super
The goal of my research to create platforms to allow modern AI configurations like Graph Neural Networks to be deployed on drones.
The goal of my research to create platforms to allow modern AI configurations like Graph Neural Networks to be deployed on drones.
Drones use computer vision machine learning techniques to navigate their environments. Modern Graph Neural Network designs enable vastly improved perception for drones, but few systems exist to deploy this work outside of simulation. The goal of this project was to create a framework based on ROS that implements the communication necessary for Graph Neural Networks. The framework was composed of several customizable programs built on an industry-standard robotics framework. Preliminary testing was conducted by running the framework with an advanced depth estimation network on drones during flight tests. Across both autonomous and human-controlled flight, message passing between two drones was successful. Computational strain and network strain were measured at manageable levels. The framework was demonstrated to be effective in real robotics applications, with successful communication occurring during drone flight. Implementation with a modern Graph Neural Network could enable real-world perception that exceeds present-day capabilities. The framework bridges the gap between modern computer science research and real-life deployment.
Press the pop-out button to view:
1st place in Engineering & Technology semifinals JSHS New York-Metro 2024
2nd place finalist in NY Metro JSHSn 2024
Oral Presenter for JSHS Nationals 2024
Terra NYC STEM Fair semifinalist 2024, finalist 2023