Avatar: A Telepresence System for the Participation in Remote Events

A unique telepresence helium blimp drone for the purpose of participating in remote events is presented. The blimp is safe for humans, can be controlled remotely over the internet, and can provide a high-quality 4K video stream to multiple users at the same time. We show that a careful selection of up-to-date hardware enables the blimp to run computationally expensive algorithms like SLAM and deep learning neural networks without relying on a local ground station to perform complex tasks. Additionally, we compare the performance of the ORB-SLAM2 algorithm when used with recent depth estimating neural networks in RGB-D mode to its performance when used with conventional passive stereo images. Our experiments show that the ORB-SLAM2 algorithm can be used with estimated depth images to reliably detect turns of the vehicle. However, in this case the algorithm also underestimates the covered distance for straight movements in one direction which results in very inaccurate computed positions.