Since the arrival of cheaper, and faster processing power and infinite amounts of storage space in 2015, research and development in artificial intelligence has boomed. They can now learn from mistakes and identify patterns from large amounts of data at a rate that is beyond the human capability. As artificial intelligence continues to make an impact in the healthcare, transportation, and education sectors, it also is making a crucial mark in the space industry.
In a spacecraft, there can be more than a 1000 variables that must be analyzed to diagnose abnormalities within the vehicle. Takehashi Yairi and other collaborators have created a data driven approach to help detect anomalies by taking in data on high dimensionality, multi-modality, heterogeneity, temporal dependence, missing data and trivial outliers. The data is then modeled to created so that anomalies are easily identified.
As we continue to expand our space exploration with unmanned vehicles, there will only be more and more of a communication delay. A rover on Mars can only be given an instruction from engineers on Earth every 20 minutes due to the communication delay. Given that only 5 commands can be given in 20 minutes, a rover can only do up to 360 commands in a day, which is highly inefficient. If the rover had an autonomous navigation capability, it could make more than 5 decisions per min. Wilkinson and Meade have proposed an artificial neural network for satellite navigation to learn the effects of engine burn through magnitude variation and direction of burn on the spacecraft’s flight path. The algorithm will continually tweak its neuron parameters until its error parameter is minimized when compared to its desired state.
Scientists are also studying planets and their geographical features to learn about their geological makeup and history. They must look through a large volume of image data sent back by on-ground rovers to Earth. If rovers had an algorithm to analyze the features and classify the images to decide its geological significance, they can then decide by themselves what the best course of action would be: disregarding the image or sending it back to Earth. Trained algorithms such as support vector machine (SVM) and continuously scaled template models (CSTM) were tested with images of Mars to detect craters. The SVM model was most successful and was as accurate as humans classifying the photos.