NASA is offering Internet users the help of identifying images captured by its rovers in order to better analyze the composition of the ground, and to apprehend potential obstacles.
A little click for man, a giant leap for humanity. For the past few days, every Internet user can help NASA continue its conquest of Mars, training algorithms to direct its Curiosity rover. The principle is simple: thanks to a platform called AI4Mars, everyone can annotate photos taken on the red planet, helping the artificial intelligence used by rovers to better understand the composition of the ground or to detect dangerous obstacles.
Called SPOC (Soil Property and Object Classification), NASA’s algorithm is responsible for classifying the elements that make up the rover exploration terrain, creating a virtual map to guide vehicles.
Identification sometimes difficult
Once on the site hosted by the collaborative portal Zooniverse, users see images appear, on which they must draw polygons in order to separate and then classify the different physical elements appearing on the screen. IN particular, NASA seeks to isolate rocks, dangerous for its vehicles, but also sandy areas, perilous for the rovers of the American institution. A sorting that can be difficult, even for specialists.
“The typology of land on Mars is often ambiguous, including for experts. You don’t need to annotate the whole image,” says the AI4Mars website, which seeks to guard against possible erroneous annotations. A directive also encourages Internet users not to overlay two polygons, in order to avoid any risk of confusion by the machine, and never to add information that they are not sure about.
“Then we hope that this algorithm can become accurate enough to help us perform other tasks, for example to predict the risk that a rover’s wheel will lack grip on a surface,” said Hiro Ono, a researcher at NASA’s Jet Propulsion Laboratory.
Currently, it takes his teams between four and five hours to plan and code a safe route to Mars, using geologists. According to NASA, artificial intelligence is not intended to replace this work but to simplify it by integrating the classification of the physical elements surrounding the vehicle. The platform, which has 8,000 images sent by Curiosity, could later be supplemented by the images of the Spirit, Opportunity and perseverance rovers, launched this summer.
The operation of the AI4Mars tool is similar to that of captcha, whichasks Internet users to prove that they are human before accessing a web page. Every day, millions of Internet users are invited to select common items (bus, fire hydrant, etc.) from different photos, to help companies like Google perfect their image recognition algorithms.