Aerospace / UAV & Drone Operations
A drone services company specializing in aerial inspection and delivery wanted to add an automatic landing system to their fleet — one that could detect whether it was safe to touch down in real-time.
One of their top concerns? Landing near people. A safe, fully autonomous system had to recognize when a person was underneath the drone and delay descent if the area wasn’t clear.
To make this possible, they needed a lightweight AI solution that could run directly on the drone — without relying on a cloud connection or large power-hungry processors.
Building such a system wasn’t straightforward. The team faced key constraints:
Limited Space & Power: The drone had strict weight and energy limits, leaving no room for large GPUs or traditional computers.
Real-Time Requirements: The system had to make instant decisions about whether it was safe to land, without delays or false positives.
No Network Reliance: Remote locations and unstable signals meant everything needed to run on-device.
We worked with the drone operator to design and deploy a compact AI solution that could detect humans beneath the drone in real-time, ensuring safer landings without compromising performance or flight time.
We trained a custom object detection model optimized for human recognition from an aerial angle — even when people were partially obscured or standing still.
The model was deployed using TensorRT, optimized for speed and minimal power usage.
It ran directly on NVIDIA Jetson TX2 and Nano boards — small, efficient AI modules ideal for UAVs — delivering fast inference without overloading the drone’s battery or CPU.
If a person was detected in the landing zone, the system instructed the drone to hover and wait, or to adjust the landing site based on predefined logic.
The final solution allowed the drone operator to:
Confidently land in mixed environments, including parks, construction sites, and remote fields.
Reduce human risk, with automated overrides that delayed landing when people were detected.
Keep performance efficient, thanks to on-device AI that didn’t require heavy hardware or constant connectivity.
This gave the drone operator a competitive edge, especially for clients requiring safety-sensitive or partially autonomous operations.
This project is a strong example of how AI can be scaled down — not just up. By fitting advanced human detection into a compact edge device, the drone operator made a leap forward in autonomous safety, opening the door for smarter, safer UAV systems across industries.
