The public may now assist NASA’s Perseverance rover in teaching an artificial intelligence program to spot scientific facts in photographs collected by the rover.
Artificial intelligence, or AI, has the potential to drastically alter how NASA’s missions investigate the cosmos. However, because all machine learning algorithms require human training, a new initiative encourages members of the public to name scientifically interesting aspects in pictures collected by NASA’s Perseverance Mars rover.
The project, dubbed AI4Mars, is a follow-up to one that was released last year and relied on pictures from NASA’s Curiosity rover. Participants in a previous round of the experiment categorized roughly half a million photographs with a tool that highlighted elements such as sand and rock that rover drivers at NASA’s Jet Propulsion Laboratory look for while planning paths on Mars. The eventual result was SPOC (Soil Property and Object Classification), an algorithm that could properly identify these traits about 98 percent of the time.
SPOC is still in development, and researchers plan to send it to Mars on a future spacecraft that will be able to do even more autonomous driving than Perseverance’s AutoNav technology.
SPOC will benefit even more from Perseverance images since the types of identifying labels that may be given to objects on the Martian surface will be expanded. AI4Mars now has labels that may be used to detect finer characteristics, such as float rocks (also known as “rock islands”) or nodules (BB-size balls, often formed by water, of minerals that have been cemented together).
The idea is to fine-tune an algorithm that will aid a future rover in finding needles in a haystack of data coming from Mars. Perseverance, which has 19 cameras, sends dozens to hundreds of photos to Earth every day for scientists and engineers to examine for particular geological traits. However, time is crucial: After those photographs have travelled millions of kilometers from Mars to Earth, the team members only have a few hours to design the next set of instructions to send to Perseverance depending on what they observe in the images.
“It’s not possible for any one scientist to look at all the downlinked images with scrutiny in such a short amount of time, every single day,”said Vivian Sun, a JPL scientist who helps oversee Perseverance’s daily operations and provided AI4Mars project consultation.
“It would save us time if there was an algorithm that could say, ‘I think I saw rock veins or nodules over here,’ and then the science team can look at those areas with more detail,” Vivian further added.
A strong dataset is essential for any successful algorithm, according to Hiro Ono, the JPL AI researcher who spearheaded the creation of AI4Mars. An algorithm learns more as more individual bits of data become accessible.
“Machine learning is very different from normal software,” Ono explained. “This isn’t like making something from scratch. Think of it as starting with a new brain. More of the effort here is getting a good dataset to teach that brain and massaging the data so it will be better learned.”
Researchers may utilize tens of thousands of photos of houses, flowers, or kittens to train their Earth-bound algorithms. However, prior to the AI4Mars initiative, there was no comparable data store for the Martian surface. The team would be happy with a store of 20,000 or more photographs, each tagged with a variety of attributes.
According to Annie Didier of JPL, who worked on the Perseverance version of AI4Mars, the Mars data repository might serve a variety of functions. “With this algorithm, the rover could automatically select science targets to drive to,” she explained. She said that it may also retain a variety of photographs aboard the rover and just transmit back images of selected characteristics that scientists are interested in.
Ono stated that the AI4Mars team believes it is critical to make their own dataset accessible so that the whole data science community may benefit.
“If someone outside JPL creates an algorithm that works better than ours using our dataset, that’s great, too,” he said. “It just makes it easier to make more discoveries.”
To assist in teaching Mars rovers how to categorize Martian topography, go to this website.