Demonstrating the Efficiency of RKO-LIO on a DJI UAV Platform Utilizing Drone Built-in IMU for GPS-Denied Mapping

Overview

This project is a lightweight 3D mapping payload for the DJI Matrice M350 RTK drone. It lets the drone build an accurate map of an outdoor area without relying on GPS, which matters around tall buildings where satellite signals become unreliable.

A LiDAR scanner and a small NVIDIA Jetson Orin NX computer ride on the drone. Instead of adding another motion sensor, the system reuses the drone's own built-in IMU, which keeps the payload cheaper, lighter and simpler to build. The whole mapping process runs onboard using ROS 2.

What the project is about

Drones often rely on GPS to know where they are, but near large buildings the signal degrades or disappears. This project gives a drone another way to understand its surroundings: it scans the environment with a LiDAR and builds a 3D map while it flies, with no help from satellites.

Comparison of the surveyed outdoor area and reconstructed coloured point cloud — The surveyed outdoor area next to the coloured 3D map the drone reconstructed.

The payload

The payload is built around a Velodyne VLP-16 LiDAR and an NVIDIA Jetson Orin NX computer, mounted on a DJI Matrice M350 RTK. Power and communication circuits were designed so the whole unit runs from the drone and talks to it over its E-Port.

The mapping payload mounted on the DJI Matrice M350 RTK during a night flight — The complete mapping payload mounted on the drone during a test flight.

How it works

Sensing: the LiDAR scans the surroundings while the drone's built-in IMU reports how it is moving.
Computing: the Jetson Orin NX combines both streams with the RKO-LIO method to track the drone and grow the 3D map in real time.
Simplicity: because the drone's own IMU is used, there is no extra sensor to buy, mount or calibrate.

Flying and mapping

The system was flown around a university building with trees, parked cars and complex facades. The drone climbed, circled and landed while the map was being built onboard.

3D flight path estimated onboard, coloured by altitude — The flight path estimated onboard, coloured by altitude.

Satellite view of the university buildings next to the reconstructed 3D point cloud — Satellite view of the site (left) next to the 3D map built from the flight (right).

Checking the map

To see whether the map can be trusted, several real building walls were measured on site with laser rangefinders and compared with the same distances in the 3D map. They agreed closely, which shows the map keeps the true proportions of the buildings.

Building facade segments used to compare real measurements with the 3D map — The building walls that were measured on site and compared with the map.

What the work covers

Hardware integration of the drone, computer and LiDAR, including power and communication circuits.
The ROS 2 software stack that connects the sensors to the mapping method.
A field flight test and evaluation of how the mapping method behaves on a drone.

The measurements, tables and extra figures from this work are in the research paper, summarised on the publication page.