Towards Autonomous Flight of Low-Cost MAVs by Using a Probabilistic Visual Odometry Approach.

In this paper we present a methodology to localise and control low-budget Micro Aerial Vehicles (MAVs) in GPS-denied environments. The control law is based on a PD controller that controls height, orientation, roll and pitch in order to enable the MAV to fly autonomously towards a specific target. The core of our approach is the implementation of a fast probabilistic approach robust to erratic motion and capable of processing imagery data transmitted from the MAV to the Ground Control Station (GCS). The latter is due to the architecture of our low-budget MAVs which can not carry out any processing on board. However, images captured with the camera mounted on board the MAV can be transmitted via either wireless LAN or through analogue transmission to the GCS, where our fast probabilistic Visual Odometry system is used in order to rapidly obtain position estimates of the vehicle. Such estimates can be used accordingly to communicate back with the vehicle in order to submit control signals to drive its autonomous flight.