Semantic Mapping in Unstructured Environments: Toward Autonomous Localization of Planetary Robotic Explorers

Highly accurate autonomous global localization of planetary robotic explorers is crucial for robust, efficient, and safe exploration of planetary bodies. Satellite-based radio-navigation systems (i.e., Global Positioning System) are unavailable in planetary applications. Thus, global localization can be achieved by relying on registration of orbital and ground imagery, using X-band Doppler radio transmissions, or direct observation of the robotic explorers in satellite imagery. While these methods have proven to be effective, they are not automated for real-time onboard use, and require significant human intervention by the ground team. This paper addresses the problem of autonomous visual global localization of planetary robotic explorers by using a trained convolutional neural network (CNN) to obtain saliency maps from semantic segmentation of ground imagery. These saliency maps are then registered to projected views of the terrain elevation maps in the rover's general region of operation, to find the optimal match that places tight constraints on the pose of the robot in a Mars body-fixed coordinate system. We provide details on the use of the DeepLab V3+ framework for semantic image segmentation of Martian landscape imagery, including fine-tune training of existing models on domain specific data. We present that the proposed method can be used for global position estimation of autonomous robots in extreme and GPS-denied environments, and provide localization performance and sensitivity analysis on a Martian landscape dataset obtained by NASA's Perseverance rover, and discuss the limitations of the method and future research directions.