Page - 8 - in Proceedings of the OAGM&ARW Joint Workshop - Vision, Automation and Robotics

Fig. 3. Drone vision-based vehicle following, marked with a 39×39 cm Apriltag. Experiment: 3 min, mean speed 7.91 km/h and top speed 13.35 km/h. b.1) Gimbal Camera Landing Platform Tracking The drone’s GPS measurements, the gimbal current orienta- tion and the camera relative pose to the marker are combined to estimate the position of the markers in world coordinates. During specific tasks, these position estimates can be used to command the gimbal to point at the marker that is positioned on top of a landing platform. This approach is used during the vehicle following, package delivery and landing tasks. b.2) Vehicle Speed Estimation The marker relative pose estimates are calculated at around 25 fps for a resolution of 1280×720 px. These estimates are stored in a queue with a length of 20 elements. The vehicle speed is estimated for every linear coordinate using linear regression on the elements of the queue, which does not incur significant computation costs. b.3) Navigation Control Algorithm The flight behavior of our drone was characterized by per- forming speed command step-response identification tests. A rough controller parameter tunning was calculated based on the resultingmodel and itwas later experimentally improved. We utilize a feedback loop controller based on the PID controller architecture for the three linear coordinates and the yaw heading. In order to improve its performance, the controller utilizes both position and speed references. The utilized measurement feedback are the position and velocity provided by the autopilot telemetry, obtained through the fusion of GPS data with the IMU and magnetometer data. III. EXPERIMENTAL RESULTS a) Package delivery mission We succeed in performing a fully autonomous mission where the drone takes-off from the truck, follows a GPS predefined flight trajectory, looks for the delivery position, proceeds to land and drop the package, takes-off again, flies back to the distribution truck, follows it for a while and lands on the static vehicle. This mission is summarized in our video1. b) Vehicle following experiment The task of the drone is to follow the vehicle that is marked with a landing-platform at a constant distance of 2.5 m from behind and above. The vehicle speed estimate, see Sec. b.2, is used as speed reference for the controller. The vehicle following experiment lasted 3 min during which the drone performed the task successfully all the time. Pictures of this experiment are shown in Fig. 3 and the logged trajectories and speeds of the drone and the vehicle are plotted in Fig. 4. Overall, during this experiment, the mean and top vehicle speed were 7.91 km/h and 13.35 km/h, and the root mean square error (RMSE) of the position and speed control tracking error were 0.37 m and 1.34 km/h. 1Package delivery demo:https://youtu.be/bxM6dls2wuo −40 −30 −20 −10 0 10 20 0 2 X (m)Y (m) drone vehicle 150 200 250 300 350 0 3.6 7.2 10.8 14.4 t (s) Fig. 4. Vehicle following experiment of 3 min duration. The plot shows the (red) drone and (blue) vehicle 3D positions and speeds over time. IV. SUMMARY In this paper we presented a fully autonomous drone that using only on-board processing is able to perform coarse navigation using GPS and vision-based precise vehicle fol- lowing and landing (see Figs. 1 & 3). Our fully autonomous package delivery flight demonstration, carried out in col- laboration with SFL Technologies, was reported by local newspapers2,3. In future work we plan to use this system as a first step towards performing autonomous landing on a moving vehicle. ACKNOWLEDGMENTS The authors thank SFL Technologies for providing the testing environment and the electric vehicle ELI [7]. REFERENCES [1] A. Agha-mohammadi, “Confidence-aware occupancy grid mapping: A planning-oriented representation of environment,” IROS2016Workshop. [2] A. Borowczyk, D.-T. Nguyen, A. P.-V. Nguyen, D. Q. Nguyen, D. Saus- sié, and J. L. Ny, “Autonomous landing of a multirotor micro air vehicle on a high velocity ground vehicle,” arXiv preprint, 2016. [3] S. Daftry, S. Zeng, A. Khan, D. Dey, N. Melik-Barkhudarov, J. A. Bagnell, and M. Hebert, “Robust monocular flight in cluttered outdoor environments,” IROS2016Workshop, arXiv:1604.04779, 2016. [4] R. D’Andrea, “Guest editorial can drones deliver?” IEEETransactions on Automation Science and Engineering, vol. 11, no. 3, 2014. [5] E. Olson, “AprilTag: A robust and flexible visual fiducial system,” in Proceedings of the IEEE International Conference on Robotics and Automation (ICRA). IEEE, May 2011, pp. 3400–3407. [6] A. Sankalp, D. Geetesh, J. Sezal, M. Daniel, A. Greg, Y. Song, and S. Sebastian, “Autonomous semantic exploration using unmanned aerial vehicles,” IEEE IROS2016Workshop, 2016. [7] SFL Technologies, “E-Mobility electric powered vehicle ELI,” http:// www.sfl-technologies.com/spektrum/e-mobility/, accessed:2017-03-14. 2ELI roll-out demo - Mein Bezirk -http://bit.ly/2hott8t 3ELI roll-out demo - Kleine Zeitung -http://bit.ly/2it6N2P 8