Flying services with a drone
summary
Built a working indoor drone mapping and delivery prototype using a Parrot AR.Drone controlled from a single Java monolith, OpenCV camera calibration from chessboard photos, and an ffmpeg-based high‑quality video stream hack. Localization uses a particle filter with sensor drift calibration and a custom mask‑based blending map that runs ~3,000 transforms/sec. First project follows a runner (successful full-course run in Amsterdam); second project localizes indoors without GPS, picks up cargo with a magnetic hook and sometimes damages glass and drones during tests.
AI-written summary · about
chapters (23)
- 0:00Intro
- 2:31Motivation
- 3:57Our approach Professional approach
- 7:58The system
- 9:30Enthusiastic approach
- 10:26Ambitions
- 11:13Manual Map Creation
- 12:27Top view blending
- 14:01Automating
- 20:36Distortions and camera paramaters
- 21:46Map created after camera calibration
- 25:50Particle filtering
- 28:15Challenges
- 29:36Reading sensors
- 32:43Modeling uncertaintity
- 33:54System architecture
- 35:26Mapping demo
- 38:13Map based control
- 41:31Conclusion
- 42:40Toilet idea
- 43:45Blending the perspectives
- 44:56Requirements
- 47:36Summary
related talks
48mFlying services with a drone
45m(Deep) Learning to Fly
48m(Deep) Learning to Fly
06mTiny Flying Robots
21mVisualizing robot minds, and other fun stuff
45mTutorial: Building a Drone, From Scratch
21mBitDrones: Towards Using 3D Nanocopter Displays as Interactive Self-Levitating Programmable Matter
28mGame of Drones