ROS2 Autonomous Explorer

Autonomous exploration robot using Next-Best-View (NBV) planning, multi-sensor fusion, SLAM, and Nav2 navigation — simulated in Ignition Gazebo 6.

Overview

A 4WD Mecanum holonomic robot that explores an unknown maze environment autonomously. It fuses LiDAR, radar, and RGB-D depth into a single scan, builds a 2D occupancy map with SLAM Toolbox, and continuously computes the next best viewpoint to maximise frontier coverage.

System Data Flow

Simulation Environment

SLAM + RViz Visualization

Project Structure

src/
├── autonomous_explorer/
│   ├── autonomous_explorer/    # Python library (nbv_utils, localization, mapping)
│   ├── scripts/                # ROS2 node executables
│   │   ├── obstacle_cluster_node.py   — LiDAR obstacle clustering
│   │   └── nbv_goal_provider_node.py  — NBV mission controller
│   ├── launch/
│   │   ├── nav2_exploration.launch.py — top-level orchestration
│   │   ├── localization.launch.py
│   │   ├── mapping.launch.py
│   │   └── navigation.launch.py
│   ├── config/                 # YAML parameters for all subsystems
│   └── urdf/                   # Robot model + Gazebo world
└── sensor_fusion/              # Multi-sensor fusion node

Quick Start

Prerequisites

• ROS2 Humble
• Ignition Gazebo 6 (Fortress)
• ros-humble-nav2-*, ros-humble-slam-toolbox, ros-humble-robot-localization

Build

cd ~/ros2-autonomous-explorer
colcon build --packages-select autonomous_explorer sensor_fusion --symlink-install
source install/setup.bash

Run

# Full system: Gazebo + SLAM + Nav2 + NBV exploration
ros2 launch autonomous_explorer nav2_exploration.launch.py

# With RViz
ros2 launch autonomous_explorer nav2_exploration.launch.py use_rviz:=true

Subsystems start in sequence:

Time	Component
T+0s	Gazebo, sensor fusion, EKF localization
T+12s	SLAM Toolbox (waits for EKF TF)
T+17s	Nav2 stack (waits for /map)
T+22s	NBV goal provider (waits for Nav2 action server)

NBV Planning Library

Core algorithms in autonomous_explorer/nbv_utils.py (pure Python, no ROS2 deps):

Class	Role
OccupancyMapper	Bayesian log-odds grid synced from SLAM Toolbox
OutlineExtractor	Polar-sector frontier detection via jump edges
CandidateGenerator	NBV candidate placement at frontiers + uniform fallback
NBVScorer	Scores candidates by visibility (ray-casting), distance, and orientation

Robot Model

4WD Mecanum holonomic base (0.5 × 0.3 m, 3.0 kg)

Sensor	Spec
Hokuyo GPU LiDAR	16-beam, 360°, 0.08–18 m, 40 Hz
9-DOF IMU	Accelerometer + gyroscope + magnetometer
RGB-D Camera	0.3–6 m depth range
Forward Radar	90° FOV, 50 m range

Tech Stack

• ROS2 Humble · Ignition Gazebo 6 · Nav2 · SLAM Toolbox
• robot_localization EKF · imu_filter_madgwick
• Python 3.10 · NumPy · SciPy