Separable Integrated Drone-Mobile Robot System

for Unmanned Patrol

Jeonghan Kim Jongchan Park Hayeon Lee Suhwan Park Yunjea Nam

Gihyeon Kwon Woohyeon Jung Youngmoon Lee

Hanyang University ERICA, Department of Robotics

ICROS 2023, June 21-23, Samcheok Sol Beach

(Separable Integrated System, Drone, Mobile Robot)

TL;DR: Carrier is a ROS-based platform where an autonomous ground robot localizes, navigates, charges, and deploys a drone for last-mile scout missions over rugged terrain.

Drones and mobile robots are useful for unmanned patrol and military scouting, but each platform has a different weakness: ground robots struggle in rugged terrain, while drones are limited by onboard battery capacity. Carrier combines both systems so a mobile robot with a larger battery can carry and charge the drone, move it near the target region, and let the drone handle terrain that blocks the ground platform.

Paper Contributions

Map-light patrol

Naver map conversion provides a global map for waypoint planning, reducing the setup burden of repeated SLAM mapping.

Carrier station

The mobile robot acts as a moving station with a docking deck and wireless charging from its larger battery.

Cooperative sensing

Robot and drone camera feeds can be sent to the control system, while Jetson Orin can run AI models such as YOLO.

Complementary mobility

The ground robot handles efficient transport and the drone covers areas that are physically difficult for wheels.

System Design

The design modules below are illustrated by the figures that follow: obstacle detection supports perception and planning, map conversion supports global planning, the hardware figure shows charging and docking, and the architecture diagrams show the localization and navigation flow.

Mobile robot localization

Wheel odometry, IMU, and RTK GNSS are fused with an Extended Kalman Filter to estimate robot pose on the map.

Environment perception

Three RealSense cameras and front/rear LiDAR support terrain and obstacle understanding; point-cloud height changes are reflected into the costmap.

Obstacle detection

Depth-camera YOLO detections and LiDAR/point-cloud data support obstacle insertion and relative-motion-aware path planning.

Global planning

Converted Naver map data is used as a global map for waypoint-based route planning, reducing wasted resources.

Wireless charging

A docking station on the mobile robot lets the drone land and charge wirelessly from the robot battery.

Drone docking

ArUco-marker station detection estimates the relative position between drone and station for corrective landing control.

Paper Figures

Obstacle detection example with marked objects
Figure 1. Obstacle detection. Related design: obstacle detection, environment perception, and local path planning. Depth-camera detections are converted into obstacle information that can be reflected in the navigation costmap.
Naver map conversion example
Figure 2. Map conversion. Related design: global planning. Naver map data is transformed into a usable planning map, reducing the need for repeated manual SLAM mapping.

Hardware System

The platform uses a commodity mobile robot base and drone, three RealSense cameras, two LiDAR sensors, a docking station, and an automatic wireless charging system for longer autonomous patrol operation.

Carrier hardware design
Hardware design. Related design: wireless charging and drone docking. The carrier deck integrates the drone landing area, wireless charging module, cameras, LiDAR, and onboard compute.

Architecture

Carrier high-level architecture
Integrated robot architecture. Related design: hardware integration. The robot combines drone hardware, ground platform, wireless charging, compute, GNSS, cameras, LiDAR, and communication modules.
Mobile robot autonomous system architecture
Mobile robot autonomous system. Related design: localization, environment perception, and navigation. Localization, indoor/outdoor navigation, local planning, environment perception, and LiDAR fusion feed the patrol stack.

ROS Stack

Carrier ROS packages
Description URDF, xacro, meshes, and RViz model configuration.
Gazebo Simulation launch files, worlds, robot model integration, and environment assets.
SLAM and Map Export GMapping launch/config files, station maps, occupancy map saving, map cropping tools, and map conversion support.
Navigation Move base, AMCL, EKF localization, indoor/outdoor parameters, RTK/GNSS-oriented launch files, and RViz navigation views.
Perception and Sensors LiDAR, IMU, three RealSense camera launch files, floor detection, sensor fusion, and environment perception.
Interfaces Battery aggregation, ArUco station detection support, web interface nodes, custom messages, and services.

Project Reference

@inproceedings{kim2023carrier,
  title = {Separable Integrated Drone-Mobile Robot System for Unmanned Patrol},
  author = {Kim, Jeonghan and Park, Jongchan and Lee, Hayeon and Park, Suhwan and Nam, Yunjea and Kwon, Gihyeon and Jung, Woohyeon and Lee, Youngmoon},
  booktitle = {ICROS 2023},
  year = {2023},
  url = {https://github.com/7drone/carrier_ros},
}