The MIT research team has developed a new approach to transform digital land maps in the space discovery. The study enables the production of DTM more sensitive and faster, even on planets where there is no GPS access, with a method based on neural volume processing. The results aim to increase the success of exploration tasks, especially in targets such as Mars.
Today, mapping on remote planets faces several basic difficulties: the stages that need to be cleaned by hand, the inconsistencies in light conditions, and the fluctuations of data quality. In order to overcome these problems, the MIT team revealed a solution called Neural Terrain Maps (NTM). This approach can produce three -dimensional maps, even if it is a defective data, by processing satellite images with advanced artificial intelligence techniques.
The main advantage of NTM is to obtain outputs faster and more reliable, even if the image quality is low. The team improved image cleaning using GAN and Transformer-based models and managed to reduce lighting problems with the Shape-From-Shaden approach. In addition, the existing Neural Radiance Fields (Nerf) technology, taking into account some limitations, the stage geometry was aimed to extract stage geometry more directly without the need for additional transactions.
The tests on the Gale Crater showed that NTM gave more successful results than Nerf. This new method provides particularly benefits on surfaces that have not previously been mapped or have data restrictions. Studies show that effective results have been achieved even in cases where traditional mapping methods are insufficient by making the cross -verification of data collected from the world and Mars with real and synthetic satellite data.
