Palm-Sized UAV Uses Omnidirectional Vision and Sparse Maps for Exploration

Classic autonomous exploration methods rely on dense occupancy maps or high-resolution point clouds, consuming significant memory and computation. This is impractical for micro UAVs constrained by size, weight, and power (SWaP). Researchers led by Zirui Wang propose a lightweight alternative: a palm-sized omnidirectional vision-based UAV guided by sparse topological maps. The drone uses a multi-fisheye camera array to achieve a full field of view and estimate depth. To overcome limited depth accuracy, frontiers (potential unexplored regions) are represented as topological nodes rather than explicit boundaries, avoiding the need for occupancy grids or global point clouds.

The system abstracts the environment into a sparse graph of key nodes with descriptors, cutting memory use and computational load dramatically. Global path planning runs directly on this graph. The method was validated both in simulation and on a real 11cm-wheelbase, 400g UAV equipped with the camera setup. Results show efficient exploration with extremely low computational overhead. This approach could enable autonomous navigation for micro-drones in applications like search-and-rescue, inspection, and environmental monitoring, where size and power are critical constraints.