1 May 2018
On a regular basis, various space debris, including sometimes threatening asteroids, comes close to or impacts the Earth. For example, in October 2017 asteroid 2012 TC4 passed by the Earth approximately 1/8th the distance between us and the moon. While this event did not present a threat, because there was no impact and because 2012 TC4 was only about 10 meters in diameter so would have burnt up in the atmosphere if it did enter, questions arise as to how to deal with asteroids that are more menacing and pose a risk of great harm. Dr. Nicolas Erasmus, an astronomer here at the South African Astronomical Observatory (SAAO), is co-author of a recently published paper that tries to address this problem.
In July 2016 Dr. Erasmus attended the Frontier Development Lab hosted by NASA’s Ames Research Center and SETI in Mountain View, California. The six-week workshop brought together planetary scientists and machine-learning experts to tackle various challenges that involve asteroids. As part of a team of four, Dr. Erasmus and his fellow team members were posed the following dilemma: “Could mankind deflect a hazardous asteroid on a crash course with Earth and if so which method would give us the best chance?”. To answer this, the team worked together to create a machine-learning algorithm called the “Deflector Selector”, which can be used to study a given population of potentially hazardous objects and then determine which technology has the best chance of deflecting them from Earth’s path.
To train the algorithm, a simulation of millions of hypothetical asteroids with the potential to hit the Earth was created. For each hypothetical impact, they simulated how far in advance of the collision it could be detected and the velocity change to the asteroid which would be required to avoid the collision with Earth. With the aid of this information, the team reviewed the capability of three technologies to induce this velocity change: a nuclear detonation, a kinetic impactor, and a gravity tractor. The technologies each work differently and present different challenges. Nuclear detonations release an explosive force that can impart momentum, which while effective, carries with it the dangers associated with launching nuclear warheads into space. The somewhat less effective kinetic impactor causes a change in momentum by crashing a spacecraft into the asteroid and is technologically the easiest method. Gravity tractors are more subtle and involve hovering a spacecraft near an asteroid, allowing its gravitational pull to nudge the asteroid in a different direction. The latter two technologies currently offer less potent results but have more predictable outcomes. Their effectiveness can also be enhanced with earlier detection and therefore longer lead times.
The benefit of using a machine learning algorithm to solve this dilemma is that while it takes a long time to generate the training data and train the algorithm with this data, it can provide clear answers extremely quickly when given a new unknown impact scenario which mankind might one day face.
- Dr. Nicolas Erasmus : firstname.lastname@example.org or 021 201 5163