Autonomous UAV Navigation System

UAV_DEMO.mp4

Dynamic A* path replanning on an evolving occupancy grid.

Autonomous UAV navigation system featuring real-time 2.5D occupancy-grid mapping, A* global path planning, and depth + LiDAR sensor fusion for dynamic obstacle avoidance in complex urban environments. Fully integrated with ROS 2, PX4 Offboard control, and Gazebo simulation for high-fidelity testing and deployment.

System Capabilities

Capability	Technical Summary
Mapping	Height-aware voxel occupancy grid updated with Bresenham ray tracing and yaw-rate filtering to prevent distortion during rotation.
Global Planning	Continuous A* replanning with safety-inflated costmap and smoothed trajectory publishing.
Local Navigation & Safety	Reactive velocity commands, corridor-centering and collision overrides based on depth and LiDAR fusion.
Flight Control	PX4 Offboard position setpoints applied through ROS 2 middleware for stable navigation.
Visualization	voxel map + global path + UAV pose in RViz for interpretability of autonomy behavior.

Directory Structure

Architecture

Functional Overview

ROS 2 Node Overview

This project is organized into clear perception, planning and control modules. Each node runs independently and communicates over ROS topics for robust modularity.

Nodes are launched together via:

ros2 launch autonomous_drone drone.launch.py

ROS Data Flow

Below is the real signal wiring used in this navigation stack:

Node Name	Purpose	Key Inputs	Key Outputs
drone.launch.py	Orchestrates the full UAV autonomy stack	Launch parameters, PX4 interface, sensor drivers	All core ROS 2 nodes running
mapping_node.py	Builds a real-time 2.5D occupancy grid of the environment	/scan, /depth_camera, /fmu/out/vehicle_local_position	/map, 3D MarkerArray
path_planner.py	Performs A* global path generation and dynamic replanning	/map	/global_path
depth_avoidance.py	Short-range sensor fusion and collision avoidance	/scan, /depth_camera	Safe proximity distance data

Real-time mapping and A*-driven navigation to user-selected waypoints in dense corridors.

Autonomous transition from reactive local avoidance to global A* navigation as mapping improves.

Data update rates:

• /depth_camera ~30 Hz
• /scan ~10 Hz
• Setpoints: 10–20 Hz
• Replanning whenever deviation or map changes

Algorithms & Techniques

Sensor Fusion (depth_avoidance.py)

• Combines depth + LiDAR with reliability weighting
• Median-based filtering to reject noise
• Safety overrides when distance < thresholds
• Generates front/left/right clearance corridors

3D Mapping (mapping_node.py)

• Ray-tracing from sensor origin to detected hit
• Voxel height stacking → vertical building shapes
• Ghost clearing to erase outdated obstacles
• Rotation-aware filtering reduces distortion at turns

Global A* Planning (path_planner.py)

• Real-time A* search over occupancy grid
• Dynamic replanning if path blocked
• Cost inflation around obstacles for clearance safety

Local Navigation (autonomous_navigator.py)

• Path tracking + lateral centering in corridors
• Adaptive forward speed based on heading alignment
• Collision stop-and-slide behavior when blocked
• Smooth yaw blending to reduce oscillation

PX4 Offboard Flight Control

• Direct setpoints in ROS → PX4 NED
• Automated arming, set-home, takeoff, landing fail-safe

Installation & Setup

Dependencies

• ROS 2 Jazzy
• PX4 Firmware (Offboard enabled)
• GZ Sim 8.10.0
• Python 3.10+

Create & Prepare Workspace

# Create a fresh ROS 2 workspace (or use your existing one)
mkdir -p ~/ros2_ws/src
cd ~/ros2_ws/src

Clone the Repository

git clone https://github.com/Ajinkya-001/autonomous-uav-navigation-system.git
cd ~/ros2_ws

Install Dependencies

sudo apt update
rosdep update
rosdep install --from-paths src --ignore-src -y

build

cd ~/px4_ros2_ws   # replace with your workspace path
colcon build --symlink-install
source install/setup.bash

Launch Simulation

ros2 launch autonomous_drone drone.launch.py

How It Works (Navigation Flow)

1. User clicks a 2D Goal Pose in RViz — sends /goal_pose.
2. UAV begins motion using local fused obstacle avoidance while the map is still unknown.
3. Once a usable map is built, the A* planner generates a global route.
4. UAV follows global path, but switches back to avoidance if blocked.
5. UAV slows down and stops once the goal is reached (<1 m proximity).

The system continuously switches between global planning and local safety behaviors depending on environmental certainty.

Configuration Parameters

Parameter	Default Value	Purpose / Effect	Tune In (Script / Location)
flight_altitude	5.0 m	Desired cruising height for waypoint tracking	drone.launch.py or inside PX4 setpoint parameters
map_resolution	0.10 m	Voxel size for occupancy grid mapping	mapping_node.py → resolution config
safety_radius	0.6 m	Clearance buffer around obstacles for A* planning	path_planner.py → A* planner parameters
replan_rate	2.0 Hz	Frequency of triggering global replanning	path_planner.py → dynamic replanning loop
base_speed	5.0–7.5 m/s	Forward motion speed based on available corridor	depth_avoidance.py → local speed logic
front_stop_threshold	1.2 m	Emergency stop threshold for sudden obstacle blocks	depth_avoidance.py → fail-safe distance check

Known Limitations

• Mapping may distort slightly during rapid yaw rotation before pose stabilization.
• Requires both LiDAR and depth camera views overlapping for best reconstruction.
• A* replanning frequency can increase in dense obstacle clusters, affecting speed.
• System currently tested only in PX4 SITL simulation (no GPS fusion enabled).
• Narrow vertical gaps may not be detected if outside depth/scan field-of-view.

Metric	Measured Result (Simulation Environment)
Navigation Success Rate	82-87% across cluttered indoor corridors (improved with lookahead control)
Average Cruise Speed	2.1 m/s during autonomous traversal (increased from 1.8 m/s with balanced tuning)
Peak Speed	Up to 9.5 m/s in open spaces (adaptive speed control)
Minimum Obstacle Clearance	0.6 m critical safety threshold, 1.8 m comfort zone
Mapping Update Rate	Depth: 30 Hz, LiDAR: 10 Hz fusion with temporal validation
Path Re-planning Frequency	2 Hz global A* updates with path hash deduplication
Control Loop Rate	10 Hz (0.1s cycle time) with acceleration limiting
Acceleration Limit	3.5 m/s² max for smooth trajectory generation
Lookahead Distance	2.5 m Pure Pursuit tracking for smoother cornering
Reliability Duration	Stable up to ~45 min continuous simulation with stuck detection
Localization Accuracy	±0.12 m positional variance from PX4 local position
Obstacle Avoidance Latency	<150 ms from detection to control action (sensor → fusion → control)
Sensor Fusion Weights	Depth: 68% (4.6 weight), LiDAR: 32% (2.17 weight) based on inverse variance
Trajectory Smoothness	30% reduction in lateral oscillation vs v5.5 (Pure Pursuit + acceleration limiting)
Stuck Detection Threshold	0.5 m movement over 3 seconds triggers waypoint skip
Failure Modes	Dead-ends (45%), sensor occlusion in tight corners (35%), map quality degradation (20%)

Future Work

• Hardware validation on real UAV (Jetson + PX4 flight controller)
• Outdoor support with GPS + IMU fusion
• Semantic layer in mapping (roads, buildings, no-fly zones)
• Moving obstacle tracking and dynamic path prediction
• Multi-UAV coordination for collaborative exploration

License

This project is licensed under the MIT License.

You are free to use, modify and distribute this software for research and development purposes.

Contact

Ajinkya Patil

B.Tech – Artificial Intelligence & Robotics

Email: ajinkyapatilckl@gmail.com

GitHub: https://github.com/Ajinkya-001