SkyDrop AI

Inspiration

In emergency situations, delivering medical supplies swiftly and safely can mean the difference between life and death. Traditional delivery methods often struggle with navigating unpredictable terrains or real-time obstacles. We were inspired to build SkyDrop AI to take the guesswork out of automated aid delivery, ensuring that first-response packages land exactly where they are needed most without human intervention or risk.

What it does

SkyDrop AI is an autonomous drone delivery system designed for precision aid drops. The core software dynamically connects to a drone's live feed using the Cyberwave API to gather real-time data.

Our system combines real-time drone imagery with Google satellite maps to create a unified, high-accuracy overhead view.
We then apply YOLO object detection to process the mixed visual data and identify obstacles or open landing zones.
This integrated analysis instantly pinpoints the safest and most accurate target location to drop the aid kit.

How we built it

We engineered SkyDrop AI using Python in a local development environment via VS Code.

Drone Integration: We utilized the Cyberwave SDK to handle autonomous flight routines (takeoff, hover, landing) and establish a direct connection to a digital twin of a DJI Mini 3 drone.
Computer Vision: We implemented the YOLO (You Only Look Once) object detection framework alongside the Pillow (PIL) library to process byte-streamed imagery from the drone camera.
Resilient Architecture: To handle the unpredictable nature of drone data streams, we built defensive error-handling loops into our flight code, ensuring the aircraft executes safe recovery and landing sequences even if a camera telemetry bug occurs.

Challenges we ran into

Transitioning from cloud-based prototypes (like Google Colab) to a local hardware-ready script introduced environment configuration hurdles, particularly around secure environment variable handling and native image rendering. Additionally, ensuring that a camera telemetry error wouldn't cause a catastrophic software freeze mid-flight was a major challenge. We overcame this by wrapping our data ingestion pipelines in strict try-except fallback layers to prioritize aircraft safety.

Accomplishments that we're proud of

We are incredibly proud of creating a robust flight loop that bridges high-level computer vision with physical drone automation. Achieving seamless image processing without compromising the safety functions of the drone—allowing it to safely complete its 10-second mission hold and land regardless of software bugs—is a massive milestone for reliable field deployment.

What we learned

We learned a great deal about the nuances of local SDK management and the realities of asynchronous communication with drone twins. This project reinforced the importance of building fail-safes into automated robotics; writing code for a drone taught us to code defensively because an unhandled exception in the real world has physical consequences.

What's next for SkyDrop AI

The next step for SkyDrop AI is integrating live physical hardware tests to evaluate how wind and varying lighting conditions affect the YOLO drop-zone detection. We also intend to implement a mechanical servo trigger hook controlled directly via the SDK, allowing the script to execute the physical drop of the medical kit the exact second the AI marks the target area clear.