Efficient Algorithms for Indoor MAV Flight Using Vision and Sonar Sensors

Abstract

This work describes an efficient perception-control coupled system and its underlying algorithms that enable autonomous exploration of indoor environments by a Micro Aerial Vehicle (MAV) equipped with a monocular camera and sonar sensors. The perception subsystem uses inputs from the camera to detect the vanishing point and doors in corridors. It detects the vanishing point by grid-based line-intersection voting (GLV) and Mixture-of-Gaussians (MoG)-based classification, while doors are detected by using simple but effective geometric scene properties (GSP) with template matching and temporal filtering. It also detects distance to obstacles, for example walls, using inputs from one forward-looking and two side-looking sonar sensors. These algorithms are accurate, computationally efficient, and suitable for real-time operation on offboard and onboard power-constrained computing platforms. The control subsystem employs a priority-based planner that combines outputs from the perception subsystem to compute high-level direction and velocity commands for the MAV. We evaluate our perception-control system on a commercially available AR.Drone 2.0 MAV with offboard processing and successfully demonstrate collision-free autonomous exploration and flight in building corridors and rooms at approximately 2 m/s speed.