• Autonomous Exploration
• Drone
• Motion Planning

In this work the challenge of autonomous exploration and mapping in underground environments using aerial robots is considered. In these environment navigation is one of the hardest difficulties to take into consideration: sensor degradation due to possible darkness, typical narrow geometries and long drifts makes commonly used exploration techniques ineffective.

The aim of this thesis is to design a motion planner for aerial robots, able to efficiently explore environments in the aforementioned conditions.

For this purpose we present a combination of Receding-Horizon Next-Best-View planning, used as Local planner, and Frontier Exploration planning, used as global planner, in order to leverage the positive features of both. In addition, a submap-based mapper is adopted: it is a kind of mapper representing the environment as a set of submaps, whose poses are optimized to find the proper alignment between them. This gives the system robustness to odometry drift.

Exploration performances results are evaluated in a simulated maze scenario with and without drift. Our apprach is compared with the state-of-the-art exploration planners using a submap-based mapper and a monolithic map: in lack of sensor noise our planner is faster at exploring than the first, but slower than the second; in presence of simulated drift our planner outperforms both, because it is still faster than the first and it completes a higher number of simulations without collisions.