Machine learning algorithms experience an increasing use in space exploration and space technology. During the 20th century, calculations for space travel were based on traditional computation methods, that take a long time, both to implement as well as to execute and that cover limited applications and scenarios. Machine learning algorithms support the space technology sector by automating and augmenting tasks such as data transmission, rocket landing, navigation, and visual data analytics.
Machine learning algorithms can for example be employed in the space tech sector for: outer space data transmission, planet data analytics, space navigation, rocket landing, rover explorations, spacecraft robotics, and space medicine.
Satellites and spacecrafts that operate in deep space collect and generate vasts amount of data. In order to analyse this data it needs to be transmitted back to earth at specific times or intervals. The data sent to and from a spacecraft can take months or even years to be transferred from the source to its destination.In case such a transmission was not successful the data could be irrevocably lost, as the data that is temporarily stored in the spacecraft will already be overwritten by the time the incomplete transmission reaches Earth.
Machine learning algorithms can be used to determine optimal ways to transfer data over long distances in space. For example, the MEXAR2 (Mars Express AI Tool) uses historic data to determine the best download schedule and optimizes the transmission of data to prevent the loss of data.
A planetary environment and condition analysis is an important phase of a deep space mission. Data about Mars was for example collected using space telescopes and satellites, before rovers were dispatched to explore the planet. Images are a major source of data. However, its interpretation and the identification of patterns in images challenging. Machine learning and computer vision algorithms can be used to support and improve the analysis of image data that was collected using telescopes and satellites.
The NASA Frontier Development Lab, collaborates with companies such as Microsoft and IBM, to develop algorithms that can detect possible solar storm damage using data that was collected by magnetospheric and atmospheric measurements. These techniques can also be used to identify suitable planet landing sites and resource discovery.
Machine learning can also be used in space exploration for the purposes of satellite motion control and spacecraft maneuvering. In order to control satellites and spaceships, large amounts of geometric and kinematic location data have to be processed in real time. As space missions become more and more common and take place at increasing distances to the earth, machine learning approaches are employed in the handling of large and complex datasets. Machine learning algorithms can be used to automate tasks such as navigation, orbit adjustment, and space station docking.
As an example of how machine learning algorithms can be employed to handle the adjustment of spacecraft motion and orientation, NASA’s Jet Propulsion Laboratory (JPL) is currently investigating novel methods that will allow spacecrafts to self-adjust and automate their velocity, orbit trajectory, and engine power. Machine learning algorithms can also be used to automate other aspects of deep space navigation.
Machine learning can be used for the design of algorithms that automate certain tasks in a rocket landing procedure. Most problems during rocket landings can be traced back to sensor and guidance issues, as well as software errors and errors with the vacuum stage. Machine learning and computer vision algorithms can be used to evaluate different techniques for a successful landing.
One of the most famous examples of using machine learning in the creation and optimization of landing systems is SpaceX’s Falcon 9. The Falcon 9 spacecraft that successfully landed in Cape Canaveral in 2015 employed computer vision algorithms and machine learning techniques. SpaceX utilized convex optimization algorithms to determine the best way to land the rocket on the platform. This application of machine learning algorithms allowed SpaceX to create the first ever reusable rocket system.
NASA has already been using machine learning algorithms for more than a decade. The Mars rovers Spirit and Opportunity that landed on Mars in 2004 employed a system called AutoNav. AutoNav is a navigation and self-driving system that is based on machine learning algorithms. The algorithms in AutoNav help the rovers navigate the rocky terrain of Mars. Curiosity, the Mars rover that landed in 2011, uses AutoNav as well.
Moreover, machine learning is used in the rovers’ AEGIS (Autonomous Exploration for Gathering Increased Science) system. The AEGIS system is responsible for the analysis of images of rock formations that might be of interest to scientists. Since the rovers are not able to send all images they capture back to earth but can only transmit selected images, the AEGIS system has to decide which images might be of interest to NASA scientists.
NASA has attempted to develop robotic astronauts for more than a decade. Recently, NASA has presented the Robonaut, a robotics system that is designed work in collaboration with human astronauts, to support them in the execution of dangerous tasks. The Robonaut vision system is based on several high-resolution cameras and convolutional neural networks. In this way, Robonauts can respond to requests created by the astronauts and can determine the appropriate course of action for the handling of each request.
As NASA aims to send astronauts father away from earth than ever before, the astronauts’ healthy of course needs to be assured. Medical facilities and tools are needed on board of the spaceships, as the astronauts will not be able to return to earth in case of a medical issue. The Exploration Medical Capability (EMC) initiative uses machine learning to engineer healthcare options based on the astronauts’ projected medical needs.
The EMC is intended to assure that astronauts remain healthy and fit during the course of their mission. It will need to be respond to hazards such as space radiation and gravitational changes, and will need to act as environment monitor on the one hand and as doctor on the other hand. The EMC will be a self-reliant medical care system that can diagnose injuries and can assist in their treatment. Of course, the EMC will be supported remotely by medical facilities and clinicians located on earth.