Vasileios Tsouvalas / Research
Robotics · Autonomous Exploration

Cooperative Aerial
& Ground Exploration.

A drone scouts beyond the known map. A ground robot turns that information into its next move.

Vasileios Tsouvalas & Kleio Baxevani

Aristotle University of Thessaloniki (AUTh) · Automation & Robotics Laboratory

Diploma thesis · 2018Distinction · 10/10Gazebo simulation
Watch demoCodeThesis (Greek)

The Idea

See Further. Choose Better.

Ground robots navigate precisely, but obstacles limit what they can see. An aerial view can reveal useful space before the ground robot travels there.

Complementary Roles

The system pairs a TurtleBot 2 ground robot with an AR.Drone 2.0. A visual marker aligns their coordinate frames. Aerial observations extend the occupancy grid, and an optional second step evaluates candidate targets from above.

The ground robot combines these observations with its own map to select and navigate to the next target. This cycle continues as the explored area grows.

Three RoboCup Rescue-inspired environments test when cooperation helps—and when the extra scouting time outweighs its benefit.

The System in Action

Original simulation · 2:29 · Silent
Mapping and exploration in the simulated environments. The silent recording uses robot views, evolving occupancy maps and trajectories to illustrate the exploration process.

The Exploration Loop

From an Aerial Observation to a Ground-Level Decision.

  1. 01

    Align

    The drone detects the visual marker on the ground robot to align the aerial and ground coordinate frames.

  2. 02

    Scout

    The drone visits the boundaries of the known map. Camera observations extend the occupancy grid beyond the ground robot’s current view.

  3. 03

    Evaluate

    The extended map produces candidate targets. In the richer strategy, the drone also inspects those targets and contributes an aerial score.

  4. 04

    Navigate

    The ground robot selects a target, plans a path and moves. New observations update the map, and the loop repeats.

Three Configurations

I · Baseline

Ground Only

The ground robot builds the map and selects targets using its own observations.

II · Cooperation

+ Aerial Scouting

The drone explores map boundaries to expand the information available for planning.

III · Cooperation

+ Target Evaluation

Aerial assessments of candidate targets supplement the expanded map.

Method: thesis chapters 3–4. Hardware was represented in Gazebo; RViz visualized maps and robot state. This page describes the simulated study.

Reported Findings

Cooperation Helps When the Extra View Earns Its Cost.

Three environments—corridors, dense obstacles and sparse obstacles—were evaluated under exploration, coverage and time-constrained objectives.

Full-Coverage Time

Minutes · Lower is better
Ground Only+ Aerial Scouting
Reported completion times for the full-coverage objective (minutes)
EnvironmentGround Only+ Aerial Scouting
Corridors3117
Dense obstacles2620
Sparse obstacles2720
Transcribed from Table 3, thesis p. 67 (PDF page 85). Values retain the table’s original rounding. Reported simulation results; uncertainty estimates are not provided.

Broader View, Faster Coverage

Aerial boundary scouting reduced reported full-coverage time in all three environments. At the 95% exploration threshold, the benefit appeared in the corridor and dense-obstacle environments; the sparse case was unchanged in Table 3.

Better Choices Have a Cost

The additional target-evaluation strategy visited fewer targets, but its aerial inspection overhead often increased completion time. Fewer targets alone do not establish a faster exploration policy.

Evidence: thesis §5.4, Tables 3–4 and comparison plots in Appendix A. Performance depends on the environment and stopping objective. The experiments were conducted in simulation.

Explore the Work

Code, Thesis and Research Context.

01 · Repository

Source code ↗

ROS controllers, simulation worlds and experiment configurations.

02 · Full report

Thesis (Greek) ↗

102 pages, including an English abstract on PDF pages 6–7.

03 · Reproduction

Environment & setup ↗

Historical dependencies, launch commands and current validation limits.

Academic Context

Degree
Diploma (Integrated MSc) in Electrical & Computer Engineering · March 2018
Recognition
Thesis distinction · 10/10
English Abstract

This thesis investigates collaboration between an unmanned aerial vehicle and an unmanned ground vehicle for autonomous exploration of unknown environments. The aerial robot localizes relative to the ground robot using a visual marker, then scouts the boundaries of the known area to extend the shared occupancy grid. A second cooperative strategy also uses aerial observations to evaluate candidate exploration targets. The ground robot selects a target, plans a path and navigates while the two agents update the map.

The system is evaluated in Gazebo across three RoboCup Rescue-inspired environments, comparing ground-only exploration with the two cooperative strategies under different exploration and coverage objectives. Aerial boundary scouting improves coverage completion in the reported experiments, while additional target evaluation reduces the number of visited targets at the cost of extra overhead. The findings illustrate how the value of cooperation depends on environmental structure and the exploration objective.

Edited English summary based on the original abstract and results; not a verbatim translation.