Overview obstacle maps for obstacle-aware navigation of autonomous drones

Area of Science:

• Robotics
• Computer Vision
• Geospatial Mapping

Background:

• Autonomous drone deployment in unknown environments is challenging.
• Lack of prior maps limits drone utility in scenarios like search and rescue.
• Onboard computation for real-time mapping is crucial for enhanced autonomy.

Purpose of the Study:

• To demonstrate the feasibility of autonomous drone deployment in unknown outdoor environments.
• To develop a vision-based system for rapid obstacle map generation using onboard computation.
• To enhance drone autonomy for tasks in unmapped areas.

Main Methods:

• A two-step vision-based mapping approach: overview flight for sparse mapping via photogrammetry, followed by georeferencing, mesh densification, and Octomap conversion.
• Utilizing onboard computation for near real-time generation of obstacle maps ($$ size in $$).
• Quantitative evaluation of map accuracy and trajectory planning, alongside experimental validation of safe navigation.

Main Results:

• Successful generation of an obstacle map in near real-time on the drone's onboard computer.
• Demonstrated feasibility of autonomous deployment and mapping in unknown outdoor environments.
• Validated accuracy of the generated map and safety of drone navigation within the mapped area.

Conclusions:

• The proposed vision-based mapping approach enables autonomous drones to rapidly create obstacle maps in unknown environments using only onboard computation.
• This system significantly enhances drone autonomy for critical applications such as search and rescue.
• The method allows sufficient time for drones to perform other tasks within the area of interest during the same mission.