Autonomous UAV Trinity Stack: AEGIS, Swarm & Zenith

ROS 2 Gazebo Level

A comprehensive, production-grade autonomous drone ecosystem. This repository integrates three core pillars of modern robotics: Failure Resilience, Multi-Agent Coordination, and AI-Driven Perception.

Architecture

To validate the VANGUARD fail-safe autonomy, we run a high-fidelity Software-in-the-Loop (SITL) simulation using PX4 Autopilot and the cutting-edge Gazebo Harmonic (GZ Sim). The ROS 2 Brain controls the drone dynamically via offboard setpoints over a high-speed Micro-XRCE-DDS bridge.

Developing an end-to-end hardware-in-the-loop simulation requires solving deep integration issues between ROS 2, Micro-XRCE-DDS, and PX4. Here are the core challenges resolved in this architecture:

1. DDS QoS Profile Mismatch (The Silent Drop):
   • Challenge: The drone armed but refused to climb. The ROS 2 publisher was set to TRANSIENT_LOCAL durability, while the PX4 Micro-XRCE-DDS subscriber strictly expected VOLATILE. The middleware silently dropped all trajectory setpoints.
   • Solution: Analyzed the DDS middleware handshake and aligned the ROS 2 QoS profile to match PX4's VOLATILE expectations, successfully restoring the high-frequency offboard control link.
2. CPU Throttling & Offboard Timeout:
   • Challenge: When screen recording software was activated, CPU spikes starved the ROS 2 node. The setpoint publish rate dropped below PX4's strict 2Hz safety cutoff, causing the drone to instantly abort flight.
   • Solution: Optimized the offboard control loop timer from 20Hz up to 50Hz, creating enough latency buffer to survive severe CPU throttling without triggering the autopilot's offboard loss failsafe.
3. Failsafe RTL Mode Rejection:
   • Challenge: Because internal pre-flight safety checks were bypassed for SITL testing, PX4 ignored standard failsafe fallback modes when a battery emergency was injected.
   • Solution: Engineered the ROS 2 safety node to explicitly inject the MAVLink VEHICLE_CMD_DO_SET_MODE for AUTO.LAND, forcing PX4 to hand over control and land gracefully without relying on internal fallbacks.
4. WSL2 Cross-Drive Compilation Bottlenecks:
   • Challenge: Compiling ROS 2 interfaces (px4_msgs) on a mounted Windows drive from WSL took over 70 minutes due to cross-OS filesystem translation overhead.
   • Solution: Re-cloned and compiled px4_msgs natively inside the WSL Linux filesystem (~/px4_ws), cutting compile time down to just 2 seconds using underlay workspace sourcing.

This repository integrates three core pillars of modern robotics: Failure Resilience, Multi-Agent Coordination, and AI-Driven Perception.

• Status: Operational ✅
• Features: Multi-state failsafe logic, high-frequency telemetry logging, and Chaos Engineering (Failure Injection) verification.

• Status: Operational ✅
• Features: Decentralized discovery via Zenoh/DDS, multi-agent occupancy grid merging, and isolated namespace scaling.

• Status: Operational ✅
• Features: Simulated YOLOv10 vision stream, reactive potential field obstacle avoidance, and VIO-ready navigation stack.

This entire stack is containerized using Docker for instant reproduction on any system.

# Clone and Build
cd Autonomous-UAV-Trinity-Stack
docker-compose up -d
# Launch the Full Trinity Stack
ros2 launch swarm_coordinator swarm_launch.py

-Autonomous-UAV-Trinity-Stack/
├── Dockerfile # Unified multi-stage container build spec
├── docker-compose.yml # Chaos-simulation local orchestration config
├── docs/
│ ├── architecture.png # High-level DDS system architecture flow
│ ├── vanguard_takeoff_hover_demo.mp4 # Takeoff offboard simulation flight video
│ └── vanguard_failsafe_landing_demo.mp4 # Battery failsafe emergency RTL flight video
├── src/
│ ├── guardian_system/ # Fail-safe monitoring logic package
│ │ ├── guardian_system/
│ │ │ ├── black_box_logger.py # High-frequency flight event logger
│ │ │ ├── guardian_node.py # Heartbeat diagnostic supervisor state machine
│ │ │ └── offboard_control.py # High-frequency PX4/DDS offboard setpoint link
│ │ └── package.xml
│ ├── swarm_coordinator/ # Multi-agent coordination package
│ │ ├── launch/
│ │ │ └── swarm_launch.py # Isolated namespace simulation launcher
│ │ ├── swarm_coordinator/
│ │ │ ├── map_merger.py # P2P map & spatial grid merging node
│ │ │ └── swarm_coordinator.py # Distributed mission sequencing node
│ │ └── package.xml
│ ├── vanguard_interfaces/ # Custom ROS 2 msg interface definitions
│ │ ├── msg/
│ │ │ ├── FailsafeAction.msg # Custom emergency action message definition
│ │ │ └── HealthStatus.msg # Custom multi-state telemetry message definition
│ │ └── CMakeLists.txt
│ └── vio_nav_stack/ # AI Perception & VIO navigation stack
│ ├── vio_nav_stack/
│ │ ├── obstacle_avoidance.py # APF reactive obstacle evasion logic
│ │ └── perception_simulator.py # YOLO object detector perception simulator
│ └── package.xml
├── zenoh_config.json5 # DDS router config for Zenoh overlay
├── LICENSE # MIT License
└── README.md # Main portfolio presentation guide

Architected and developed as a Senior Autonomous Systems Portfolio by Yogesh.