Semantic Segmentation on Martian Terrain for Navigation using Transformers

For the longest time, it has been known that Earth is the only sustainable human-living planet across the universe. Spacecraft have not discovered the existence of life outside of our solar system. Given Mars’ proximity and similarities to Earth, the possibility of life there is a subject of interest. Scientists have been striving to learn more about Mars for decades. At this time, it is physically impossible for humans to spend an extended period of time on Mars to study its habitat. This requires relying on extraterrestrial machinery to serve as our eyes. It involves giving machines the best training available so that they can learn how to navigate the terrain. These machines capture and send forth data which is processed to morph into mapping the sphere, forecasting paths, analyzing the landscape and even improving the same spacecraft’s capabilities on learning more about the planet it is on.The rovers’ in-situ landscape images are the main subject of this study. In this research, a novel approach has been presented for multi-label semantic segmentation of the martian terrain for navigation using variants of a transformer model - SegFormer, which has resulted in a commendable accuracy of 90.86% and mIoU of 83.55% on the AI4Mars dataset.