The Age of Discovery began in the 15th century, when Europeans built their first oceangoing vessels and set out to explore the world. Whether motivated by political, economic or cultural factors, human exploration has traditionally been driven by technological progress.
Rocket booster technology developed during World War II enabled the first generation of spaceflight in the mid 20th-century, when the Soviet Union and the United States launched artificial satellites and interplanetary probes. As humans step up to deep space exploration, artificial intelligence technologies are expected to play a huge role.
Machine Learning and Space Exploration
The “learning” part of machine learning refers to an algorithm’s ability to find patterns in data to self-improve the machine’s outcomes, ie to use existing data to predict unknowns. Machine learning already has applications in banking, healthcare, aviation, and so on, and the technology is expected to power future space exploration as it can handle huge data volumes, find patterns in planet image datasets, and predict spaceship condition.
The role of machine learning in space exploration can be roughly divided into data transmission, visual data analytics, navigation, and rocket landing.
Machine Learning In Space Exploration
Machine Learning in Outer Space Data Transmission
Spacecraft and satellites operating in deep space can generate huge amounts of data due to the complexity of their research missions. Because of the different rotations and orbits of their host planets, these massive data packets must be transmitted to earth during specific windows of opportunity. The lag meanwhile will depend on Earth’s light year distance from the spacecraft’s host planet and may be months or even years. Moreover if a data packet transmission is unsuccessful, the data may be permanently lost if it was overwritten with new data in the onboard memory.
Machine learning enables a “smart” method to manage the distant planet to Earth data transmission problem. The outer space machine learning application MEXAR2 (‘Mars Express AI Tool) was introduced in 2005 at Italy’s Institute for Cognitive Science and Technology (ISTC-CNR). The onboard learning algorithm can leverage historical data to remove superfluous data and pinpoint the download schedule to optimize data packet transmission. This outer data transmission technique is already being used by NASA and others in their space research programs.
Machine Learning in Planet Data Analytics
A usual early step in deep space exploration is planet condition and environment analysis. Satellites and space telescopes have already collected a large amount data for example for target planet Mars. Images are the major data source, while the major challenge is how to identify and read the right information from the images. Machine learning has become an effective technique for solving this problem.
The NASA Frontier Development Lab and top-tier technologies companies such as IBM and Microsoft are collaborating on machine learning as a solution for solar storm damage detection, targeting a target planet’s ‘space weather’ through magnetosphere and atmosphere measurement. The technique can also be used for resource discovery and to identify suitable planet landing sites.
Machine Learning in Space Navigation
Another field where machine learning can improve current technology is in relative spacecraft and satellite motion control. Each control action selected for spaceships or satellites requires considering and processing geometric and kinematical location information in an extremely short timeframe. As outer space missions become increasingly frequent and complex and spacecraft get further from Earth, there will be growing demand for fast and self-adjusting machine-learning based navigation capabilities. The field could include orbit adjustment, autonomous navigation, and space station docking.
The NASA Jet Propulsion Laboratory (JPL) is already involved in the above research field, and machine learning has emerged as a key technique for measurement and adjustment of a spacecraft’s motion with different orbital parameters. This allows the spacecraft to self-adjust for example orbit and velocity, and can support ground navigation systems to control a spacecraft’s flight path, engine power and orbital position. A spacecraft’s onboard machine learning algorithm also has the potential to perform autonomous navigation in deep space.
Machine Learning in Rocket Landing
Recent research in landing spacecraft has focused on developing algorithms that increase the level of autonomy for air and space systems. Some of the major issues for spaceship or rocket landings include vacuum stage, software errors, guidance and sensor problems etc. Machine learning and computer vision are the core optimization and evaluation techniques for successful landings.
The SpaceX Falcon 9’s successful landing at Cape Canaveral Air Force Station in 2015 demonstrated machine learning and computer vision’s power to transform space exploration. SpaceX used a convex optimization algorithm to determine the best way to land the rocket, with real-time computer vision data aiding route prediction. These advanced machine learning applications enabled the first reusable rocket in space exploration history — a feat scientists regard as essential in developing deep space exploration.
Next Generation Space Exploration
Despite the challenges, machine learning will, or must, play a vital role in the coming age of space exploration.
With the help of advanced machine learning based terrain classifiers and path planning algorithms, NASA built a Mars Rover which can navigate long distances on the planet’s complex surface. Mars may be humans’ current target but the red planet will not be our final destination. There are reports that NASA will deploy robotic machine learning based probes to Jupiter’s moon Europa to search for life, and NASA engineer Hiro Ono says autonomous spacecraft are in the design phase.
Soon, spacecraft may operate using only artificial intelligence and machine learning algorithms. As in the past, it is technological innovations that will enable humans to go “where no man has gone before.” For the immediate future, those innovations will continue to emerge from machine learning.
Analyst: Robert Tian| Editor: Michael Sarazen