Kavach Project

Kavach is an experimental autonomous UAV system focused on real-time perception, decision-making, and advanced control maneuvers using PX4-based flight systems and MAVSDK. The project aims to develop autonomous drone behaviors, including high-speed dives, VTOL transitions, and vision-based targeting.

---

🚀 Key Capabilities

• Autonomous Dive Missions: High-speed ballistic dives for fixed-wing and VTOL aircraft.
• VTOL Operations: Analysis and implementation of transition logic (Multicopter $\leftrightarrow$ Fixed-Wing) and hover stability.
• Computer Vision Integration:
  • Real-time object detection (Color-based, YOLO).
  • Integration with Gazebo camera sensors (Depth, RGB).
  • Custom Python scripts for image processing (gazebo_camera_poc.py).
• Custom Simulation Models:
  • vtol_downward_depth_camera: A standard VTOL with a 70° downward-pitched depth camera.
  • depth_camera: Standalone sensor model.

🛠️ Tech Stack

• Flight Control: PX4 Autopilot (v1.14+)
• Simulation: Gazebo Classic (Simulated SITL)
• SDK: MAVSDK-Python
• Language: Python 3.10+, Bash
• Vision: OpenCV, NumPy

---

📂 Project Structure

Kavach/
├── Diving_Scripts/                  # Core autonomous mission scripts
│   ├── fixedwing/                   # Plans & Dive logic for fixed-wing/VTOL
│   │   ├── mission_aggressive_dive_optimized.py  # Optimized dive logic
│   │   ├── mission_dive_vtol_camera.py           # VTOL-specific dive with camera
│   │   ├── mission_dive_pinpoint.py              # Precision dive targeting
│   │   └── ...
│   ├── multicopter/                 # Quadcopter-specific missions
│   │   └── mission_dive.py
│   └── logs/                        # Telemetry and flight logs
├── models/                          # Custom Gazebo/SDF models
├── launch_vtol_large_world.sh       # Main simulation launcher
├── gazebo_camera_poc.py             # Camera Proof-of-Concept script
└── ...

---

⚡ Quick Start

1. Launch Simulation

The primary launch script initializes the PX4 SITL environment with the custom VTOL model in Gazebo.

cd /home/udbhaw/Kavach
./launch_vtol_large_world.sh

This script will:

• Kill existing PX4/Gazebo instances.
• Build and launch px4_sitl with gazebo-classic_vtol_downward_depth_camera.

2. Run Mission Scripts

Open a new terminal to run the mission scripts.

VTOL Camera Dive:

python3 Diving_Scripts/fixedwing/mission_dive_vtol_camera.py

Fixed-Wing Aggressive Dive:

python3 Diving_Scripts/fixedwing/mission_aggressive_dive_optimized.py

3. Run Camera PoC

To verify the camera feed and object detection:

python3 gazebo_camera_poc.py

---

📝 Recent Developments & Workflow

1. VTOL Camera Configuration

• Model: Modified vtol_downward_depth_camera to include a depth camera pitched at 70° (1.22 rad) for better ground visibility during forward flight.
• Status: Successfully integrated and visible in Gazebo.

2. Dive Mission Logic

• Fixed-Wing: Debugged altitude hold issues and optimized dive parameters (FW_T_CLMB_MAX, FW_T_SINK_MAX) for steeper, faster descents.
• VTOL: Implemented transition logic to handle COMMAND_DENIED errors by ensuring proper mode switching (Hold -> Stabilized -> Mission).

3. Vision System

• PoC: Established a pipeline to read raw image data from Gazebo topics via pygazebo and process it with OpenCV.
• Detection: Basic color tracking implemented as a proof of concept.

---

⚠️ Common Issues & Fixes

• COMMAND_DENIED during Transition: Ensure the vehicle is armed and in a stable mode (e.g., Hold/Position) before commanding a transition. The scripts include retry logic for this.
• Gazebo Path Issues: If models are missing, run source Tools/simulation/gazebo-classic/setup_gazebo.bash in the PX4 directory or check GAZEBO_MODEL_PATH.

---

🗓️ Future Works

• Integrate YOLOv8 for advanced object detection.
• Refine terminal phase guidance for dive missions.
• Hardware-in-the-Loop (HITL) testing.