Skip to content
Tyler Jackson
All projects
Project Kestrel logo.

// PROJECT · Handed off · Jun 2024 – Aug 2025

Kestrel

Autonomous videography drone built by a student team, sponsored by Blue Origin.

Project Lead / Pathing Team Lead · ACM at UCF · Sponsored by Blue Origin

mAP
23.64%mAP
Rank-1 accuracy
35.34%Rank-1 accuracy
Rank-5 accuracy
48.99%Rank-5 accuracy
Best-case flight time
40 minBest-case flight time
Maximum speed
40 mphMaximum speed

Overview

An autonomous drone project leveraging modern robotics, computer vision, and embedded systems to achieve intelligent flight and person-tracking capabilities using a modular ROS 2 stack and on-board edge AI computing.

What I did

  • Defined system architecture and data pipelines across sensing, perception, and control; applied systems engineering practice to coordinate subteams and delivery.
  • Designed a MobileNetV3 CNN integrated into a DeepSORT model, achieving 23.64% mAP, 35.34% Rank-1 Accuracy, and 48.99% Rank-5 Accuracy.
  • Designed and simulated the custom Kestrel drone, achieving a best-case flight time of 40 min and a maximum speed of 40 mph. Sensor suite included a camera on a dual-pivot mount, 16 short-range ToF sensors, and 4 long-range LiDAR sensors for obstacle detection.

Architecture

  • Modular ROS 2 Jazzy stack across perception, path planning, flight control, and base-station communication.
  • D* Lite planner over layered costmaps.
  • YOLO detection with DeepSORT tracking and a MobileNetV3 triplet-loss re-identification model running on a Jetson Nano.
  • ArduPilot with MAVLink and a Pixhawk flight controller.
  • Gazebo and SITL simulation for path planning and obstacle avoidance testing.

Stack

  • ROS 2
  • Python
  • C++
  • ArduPilot
  • MAVLink
  • Gazebo
  • YOLO
  • DeepSORT
  • MobileNetV3
  • Jetson Nano
  • Docker

Timeline

  1. Kestrel Test Flights

    These test flights showcased the DeepSORT system for object tracking, the obstacle avoidance system and flight pathing system, and finally the base station communication. Everything was successful and showed a promising wrap-up for the project.

  2. Project Handoff

    As I began my capstone project, VITALS, I chose to hand off the project to ACM and the leadership I put in position to continue the implementation of the research and continue working towards test flights and wrapping up the project.

  3. Finalize Project Structure and Roles

    Moving into the summer of 2025, the project underwent restructuring within the leadership and the structure of the club support. Before 2025, we were mainly supported by the ACM club as well as other smaller clubs. With the help of some dedicated members within ACM, we were able to acquire a partnership within the IEEE club at UCF, acquiring many skilled Electrical and Computer Engineers to help advance the project in aspects that were lacking.

    Restructure documents (Google Drive)

  4. Blue Origin sponsorship

    Applied for Blue Origin Technical Project Sponsorship and was rewarded a 1000 dollar sponsorship as well as technical mentorship to help advance the project forward.

    Sponsorship award (Google Drive)

  5. Project kickoff

    This was the first meeting for Project Kestrel. This Meeting was held through the Association for Computing Machinery (ACM) club at UCF, as well as publicized through the AI@UCF club, and later on partnered with several other clubs. In this meeting, we reviewed TensorFlow's introductory YouTube videos on how CNN's work and provided a simple introduction to the project's goal, vision, and timeline.

    About twenty Project Kestrel members standing together in a classroom in front of a projected slide reading 'Welcome to Kestrel UCF'.

    Kickoff photos (Google Drive)

Gallery

  • Top-down view of the black Kestrel BoB v1 circuit board, co-branded IEEE UCF and ACM, bolted to a red-and-white quadcopter frame among motor wiring and electronic speed controllers.
    The Kestrel BoB v1 board mounted on the airframe.
  • The assembled Kestrel quadcopter on a white workbench, with red arms, four black propellers, printed landing legs, and a downward-facing camera module.
    The assembled Kestrel quadcopter.
  • About twenty Project Kestrel members standing together in a classroom in front of a projected slide reading 'Welcome to Kestrel UCF'.
    The Project Kestrel team.
  • Five Kestrel team leads standing in front of a projected Kestrel UCF website prototype.
    Kestrel team leads.
  • Students seated at classroom tables watching a projected TensorFlow slide reading 'Machine learning zero to hero — Part 1: Intro to ML' at the first Project Kestrel meeting.
    The first Project Kestrel meeting.
  • Kestrel members working on laptops around classroom tables, with a numbered agenda and system sketches on the whiteboards behind them.
    A working meeting.

Documents & video

Links

SourceWebsite repo