Autonomous Aerial Vehicle for GPS-Denied Environments
As part of a practicum project under ISRO IROC-25, I led the development of an Autonomous Aerial Vehicle equipped for advanced 3D navigation and mapping. The drone was designed to operate in GPS-denied environments, with a focus on obstacle avoidance, autonomous landing, and safe return protocols.
This project represents a significant contribution to ISRO's initiative on indoor autonomous vehicles for research and rescue operations, showcasing cutting-edge computer vision and robotics integration.
Utilized the OAK-D Lite 3D stereo depth camera to generate real-time depth maps of the environment, enabling precise spatial awareness and navigation in complex indoor spaces.
Implemented advanced obstacle detection using depth frames and image segmentation to identify both dynamic and static obstacles, ensuring safe autonomous flight.
Built comprehensive navigation logic in ROS2 Foxy, integrating path planning, object avoidance, and real-time feedback loops for seamless autonomous operation.
Designed a state-based fallback system to autonomously return the drone to its starting location in case of signal loss or low battery conditions.
Applied contour analysis and variance checks on elevation maps to identify flat, obstacle-free zones for safe autonomous landing operations.
Developed a fully functional testbed using Gazebo simulator for aerial dynamics testing before hardware deployment, ensuring system reliability.
The system architecture was built around ROS2 Foxy as the core middleware, providing robust inter-process communication and modular design. The navigation stack integrated multiple subsystems including sensor fusion, path planning, and control algorithms.
Key technical challenges included handling real-time depth processing at 30fps while maintaining computational efficiency on embedded hardware, and developing robust state machines for autonomous decision-making in unpredictable environments.
The project successfully demonstrated full aerial coverage of an indoor arena with high spatial accuracy, proving the viability of GPS-denied navigation systems for real-world applications.
This work directly contributed to ISRO's research initiatives on autonomous vehicles for search and rescue operations, providing a robust foundation for future development of indoor navigation systems.
The simulation-first approach enabled rapid prototyping and testing, reducing development time by approximately 40% compared to traditional hardware-first methodologies.