FusionLoc: Camera-2D LiDAR Fusion Using Multi-Head Self-Attention for End-to-End Serving Robot Relocalization

As technology advances in autonomous mobile robots, service robots have been actively used more and for various purposes. Especially, serving not surprising products anymore since the COVID-19 pandemic. One of practical problems operating a robot is that it often fails to estimate its pose on map moves around. Whenever failure happens, servers should bring initial location reboot manually. In this paper, we focus end-to-end relocalization address problem. It predict directly from only onboard sensor data using neural networks. particular, propose deep network architecture based camera-2D LiDAR fusion. We call proposed method FusionLoc. method, multi-head self-attention complements different types information captured by two sensors regress pose. Our experiments dataset collected commercial demonstrate FusionLoc can provide better performances than previous methods taking single image or 2D point cloud as well straightforward fusion concatenating their features.