Distant planetary rovers communicate with scientists only intermittently because of limited visibility and availability of Earth-based antennas. As the number of spacecraft needing attention grows and spacecraft longevity increases, this bandwidth constraint will only get worse. It is therefore critical to maximize the information content in each transmission. Autonomy can improve productivity in intervals between communication opportunities and increase science data returned. In particular, science autonomy makes decisions affecting the scientific measurements that will be collected or transmitted. There is a need for smart software that intelligently targets, acquires, analyzes, and compresses imaging spectroscopy data. We will to apply the state of art in science autonomy, including the recent work at Carnegie Mellon in areas of automatic spectrometer targeting and spectra collection, science-guided path planning, and orbital terrain classification, to the creation of Intelligent Spectrometry for Robotic Explorers (ISRE). ISRE will enable real-time analysis of spectroscopic data to guide spectrometer targeting. This involves selecting rover navigation goals and directing a spectrometer foreoptic to measure the intended target rock or soil. In this way, the most informative science targets will be automatically sampled, the quality of the science data return will improve, abd the required scientist effort and necessary communication bandwidth will be reduced.
There are significant commercial opportunities for technologies like ISRE that can automatically classify and characterize the classes within large satellite data sets. Many applications in geology, hydrology, and agriculture could benefit from the advanced machine-learning techniques that ISRE will employ. In particular, we believe that ISRE represents a novel and commercially lucrative method of extracting information from satellite data in the following three ways: (1) The use of unsupervised learning dramatically reduces the manual effort required to train the information-extraction system; (2) Our planning algorithms can decide when and where to collect follow-up data (e.g,. by pointing a satellite or taking ground-truth measurements), which are typically expensive to collect, and so they should only be requested on the most useful samples -- a direct analogy to the challenge of returning science data from the surface of Mars; (4) ISRE's is designed to detect outlier data that fails to be well classified. This capability can focus the analysis of geologists onto map features or regions that are poorly understood.
More »