Final Test Flight

An autonomous flight system built around NVIDIA’s Jetson Nano running YOLOv3 for real-time object detection and navigation.

Project Overview

During my time at the University of Minnesota Duluth, I built an autonomous octocopter that combined embedded AI with flight control systems. The goal was to create a drone capable of autonomous navigation using real-time computer vision.

Hardware Architecture

Core Components

Flight Platform

AI Computing

Sensors & Connectivity

Hardware Integration Photos

Jetson Nano Bottom View 1 Bottom view showing the NVIDIA Jetson Nano mounted on the octocopter frame with the Pixhawk flight controller, GoPro camera, and power distribution system

Jetson Nano Bottom View 2 Close-up of the Jetson Nano with custom cooling fan, showing the USB ports, GPIO header, and power connections

Integration Challenges

Integrating the Jetson Nano with the flight controller required:

YOLOv3 Real-Time Object Detection

Why YOLOv3?

YOLOv3 (You Only Look Once) was chosen for its balance of accuracy and speed - critical for real-time autonomous navigation:

Implementation Details

Model Optimization

Real-Time Pipeline

  1. Capture frame from camera
  2. Preprocess and resize (416x416)
  3. Run YOLOv3 inference on GPU
  4. Parse detections and compute bounding boxes
  5. Send navigation commands to flight controller

Autonomous Navigation

The system operated in several modes:

Object Tracking Mode

Waypoint Navigation

Landing Assistance

Results & Performance

Winter Flight Testing Field testing in winter conditions - the octocopter proved reliable even in cold weather and challenging lighting conditions

What Worked

Flight Stability

Detection Performance

Autonomous Behavior

Limitations

Power Constraints

Processing Latency

Weather Sensitivity

Technical Lessons Learned

Embedded AI Optimization

Hardware Integration

Autonomous Systems

Future Improvements

If I were to rebuild this system today, I would:

Hardware Upgrades

Software Enhancements

Safety Features

Conclusion

This project was an excellent introduction to autonomous robotics, combining:

The experience of getting AI running efficiently on embedded hardware while meeting the strict timing requirements of autonomous flight taught valuable lessons about system integration and real-world constraints.

While commercial solutions have since surpassed this custom build, the hands-on experience of building an autonomous system from scratch provided insights that aren’t available from using off-the-shelf solutions.

Technical Stack


This project was completed in 2018 as part of my studies at the University of Minnesota Duluth.