Efficient RGBD-SLAM for Autonomous Micro Aerial Vehicles

Sebastian A. Scherer

RGBD Cameras

RGBD Cameras provide both regular color (RGB) and depth(D) images at the same time. With known camera calibration parameters, this allows the depth image to be warped such that corresponding color and depth data are at the same pixel location.
Fig. 1 A typical RGBD image pair captured by the camera mounted on our MAV while flying in our laboratory.

Micro Aerial Vehicle

We use a Mikrokopter quadrotor helicopter controlled by a pxIMU autopilot. The onboard computer is a COM express single-board computer (SBC) hosted by a pxCOMex base board. Its main sensor is an ASUS Xtion Pro Live RGBD camera.
Fig. 2 The Micro Aerial Vehicle (MAV) used in this work.

SLAM Method

We extend keyframe- and keypoint-based monocular visual SLAM by considering a combination of 2D and depth reprojection errors [4]. This leads to metrically correct and accurate localization and mapping results and works even for rather sparse depth images. Since we use depth measurements at locations of keypoints only, our approach is also applicable to sparse stereo vision [3]. Optimizing the map using bundle adjustment alone grows computationally too expensive very quickly and global optimization is not feasible in real-time even for small map sizes. We therefore employ a combination of local bundle adjustment and pose graph optimization to enable SLAM handling closed loops, running in real-time on our MAV[1].

Results

Our method allows localization at camera rate (30 Hz) and building maps in real-time on computationally limited hardware. Even though this work is mainly aimed at MAVs, it can also be employed to build bigger maps using rolling robots.
Fig. 3 Visualization of the maps built in real-time on the onboard computer while flying in our laboratory
Fig. 4 Map built by a robot driving through the library of our department.


Contact

Sebastian A. Scherer
Tel.: +49 7071 29 78989
sebastian.scherer at uni-tuebingen.de

Publications

[1] Sebastian A. Scherer, Shaowu Yang, and Andreas Zell. Dctam: Drift-corrected tracking and mapping for autonomous micro aerial vehicles. In Unmanned Aircraft Systems (ICUAS), 2015 International Conference on, pages 1094-1101, Denver, CO, USA, June 2015. [ details | link | pdf ]
[2] Sebastian A. Scherer and Andreas Zell. Efficient Onboard RGBD-SLAM for Fully Autonomous MAVs. In IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2013), Tokyo Big Sight, Japan, November 2013. [ details | pdf ]
[3] Sebastian A. Scherer, Alina Kloss, and Andreas Zell. Loop Closure Detection using Depth Images. In European Conference on Mobile Robots (ECMR 2013), Barcelona, Catalonia, Spain, September 2013. [ details | pdf ]
[4] Konstantin Schauwecker, Nan Rosemary Ke, Sebastian A. Scherer, and Andreas Zell. Markerless Visual Control of a Quad-Rotor Micro Aerial Vehicle by Means of On-Board Stereo Processing. In 22nd Conference on Autonomous Mobile Systems (AMS), pages 11-20, Stuttgart, Germany, September 2012. Springer. [ link | pdf ]
[5] Sebastian A. Scherer, Daniel Dube, and Andreas Zell. Using Depth in Visual Simultaneous Localisation and Mapping. In IEEE International Conference on Robotics and Automation, St. Paul, Minnesota, USA, May 2012. [ details | pdf ]

Export list of publications as BibTeX / Endnote / Rich Text Format (RTF)