sFly: Autonomous MAV Navigation Using On-Board Computed Monocular SLAM

In this video we show autonomous flights of a UAV in unknown environments. We use a key-frame based monocular SLAM as pose estimator to feed the MAV controller. Only one single camera and one IMU is used as sensors. All calculations are done on-board. To keep the calculation cost at constant complexity we delete old key-frames and features of the SLAM map. Thus we only keep record of the past few meters we flew.
The first part of the video shows an indoor flight (about 50m). Note that the square flown is not entirely mapped. Merely one edge of the square is in the current map at any given time-step. That is, the MAV perceives each step as completely new terrain. This way the monocular SLAM may be seen as "degenerated" to visual odometry.
The second part demonstrates the same approach outdoors for a trajectory of about 25m.
This video is played 2x the normal speed.

Category:

Science & Technology

Tags:

• MAV