Drone Autonomy in ROS 2 🚁

A complete ROS 2 autonomous drone navigation framework utilizing ArduPilot, MAVROS, and Nav2 for real-time path planning and dynamic obstacle avoidance.

🌟 Key Features

• ROS 2 Integration: Fully built on the ROS 2 ecosystem.
• ArduCopter & MAVROS: Seamless flight control and telemetry streaming to offboard navigation nodes.
• Nav2 Autonomy Stack: Real-time path planning globally and locally using the Nav2 framework.
• Dynamic Obstacle Avoidance: Utilizes a LiDAR (or virtual_lidar_node in SITL) with Nav2 costmaps to dynamically dodge obstacles during flight.

🏗️ Hardware & Software Architecture Diagram

graph TD
    subgraph Hardware Layer
        Lidar[Slamtec RPLiDAR C1M1-R2]
        CC[Companion Computer <br> Raspberry Pi 5]
        NPU[AI Accelerator <br> Hailo 8 AI Hat]
        FC[Flight Controller <br> Pixhawk 2.4.8]
    end

    subgraph "ROS 2 Software Stack (Companion Computer)"
        FakeLidar[virtual_lidar_node.py]
        Nav2[Nav2 Costmap & Planners]
        ObstacleNode[nav2_obstacle_node.py]
        MavrosNode[MAVROS Node]
    end

    %% Connections
    Lidar -->|LaserScan| Nav2
    NPU -.->|Semantic Vision / Future| ObstacleNode
    CC <--> NPU
    FakeLidar -.->|Simulated Scan for testing| Nav2
    Nav2 -->|Planned Path / Twist| ObstacleNode
    ObstacleNode -->|cmd_vel| MavrosNode
    
    Lidar <-->|USB / I2C| CC
    MavrosNode <-->|Telem2 UART MAVLink| FC

Loading

🧠 Obstacle Avoidance Algorithm & Sensor Setup

Sensor Setup: 2D 360-Degree LiDAR

The system currently relies on a 2D 360-degree LiDAR (such as the RPLiDAR C1) spinning at a fixed rate, which outputs a /scan topic covering all directions in a flat horizontal plane. The Nav2 framework takes this point cloud and maps it onto a real-time, cell-based 2D Costmap around the drone, artificially inflating the walls to create a buffer zone.

Custom Reactive Ego-Centric Sector Algorithm

Instead of standard path planning formulas (A*, DWB), this project uses a bespoke, high-performance logic optimized for ArduPilot's physics:

1. Grid Extraction: nav2_obstacle_node.py extracts a precise 100x100 cell window (representing the immediate flight zone) from the Nav2 Global Costmap.
2. 4-Way Sector Division: The window is split into four ego-centric zones (Front, Left, Right, Back) relative to the drone's nose.
3. Danger Evaluation: If the 'Front' sector cost exceeds a calibrated safety threshold, the drone stops waypoint tracking.
4. Reactive Escape: It compares the 'Left' and 'Right' sector densities and injects high-priority MAVROS velocity (Twist) commands targeting the clearest path.
5. 360-Degree Trapping: If the Front, Left, and Right are all blocked, it verifies the 'Back' sector. If the back is clear, it dynamically reverses out of the dead end. If all 4 sectors are blocked (Full 360-degree trap), it executes a vertical Twist escape, climbing in altitude to fly over the dynamic trap!