Receding Horizon "Next-Best-View" Planner for 3D Exploration

Proposes a path planner for autonomous exploration and mapping of unknown space by aerial robots. It grows a random tree online, picks the branch that would observe the most unmapped space, executes only its first edge and repeats, eventually exploring the whole space. Designed for limited onboard computing, it was validated in simulation and on a real micro aerial vehicle, with complexity analysis showing it scales to large, complex settings.

Key findings

• Receding-horizon next-best-view: grow random tree, take first edge of the most informative branch, repeat
• Runs online on a resource-limited micro aerial vehicle
• Computational complexity scales well to large environments

Limitations

• Abstract only reviewed; abstract gives no numeric results
• Exploration of unknown space, not optimisation of inspection image quality for a known structure