Though we may not have found intelligent life on Mars, NASA has just beamed up its own.
As announced at the end of March, NASA’s Jet Propulsion Laboratories has upgraded the Opportunity rover (already stationed on Mars) with artificial intelligence firmware, code-named AEGIS. Short for Autonomous Exploration for Gathering Increased Science, AEGIS allows the Opportunity to identify high-value photography targets — making its own decisions about which Martian rocks to photograph and send back to Earth. As the rover has limited downlink capacity, this is expected to greatly increase its productivity, allowing it to retrieve more data in fewer trips across Mars’ surface. AEGIS isn’t the first artificial intelligence application developed for space, or even at Jet Propulsion Labs — JPL has been in the game as far back as the Deep Space 1 craft in 1998.
I visited JPL on a recent rainy afternoon. Nestled in the mountains near Pasadena, California, the NASA campus dates to the 1940s, and was an early stalwart of the United States’ rocketry and space programs. Beyond security checkpoints, rows of polished, glass-and-steel buildings house the facility’s various projects — major foci at the moment are the Mars rovers and Reconnaissance orbiter, the Cassini-Huygens mission to Saturn, and the Spitzer space telescope. Further up the hill is a simulated outdoor Martian landscape, with volcanic rocks resting in red sand. It’s an eerie thing to see through a gray LA fog.
Guy Webster, my contact in JPL’s press department, gave me a tour of the rover facilities, including labs with men in white lab coats and masks (sterilized in order to keep Earth bacteria from infecting Mars) working on rover construction — from the rovers themselves to the two-part, turtle shell-like casings they are deployed in; hard cover on top and heat shield on the bottom.
My ultimate destination was JPL’s Artificial Intelligence and Machine Learning facilities. While small compared to the rest of the campus, the Machine Learning labs are playing an increasingly prominent role in space design. Composed primarily of younger people (the average age of scientists is early-to-mid 30s) the tightly knit staff of the Machine Learning department spends much of its day in close collaboration. The department has multiple open laboratories where testing is conducted on solving rover logistical problems using AI. In one area, a Martian sandpit has been constructed and a Spirit-model rover placed within, built to mirror a similar situation on Mars itself, where the actual Spirit rover has become stuck. The team has so far been unsuccessful at extricating the rover from the sandpit using software, though the learning process continues.
The AEGIS software was developed here and then sent to three satellite transmitter stations in three different spots on the globe (in the Mojave Desert, Canberra and Madrid), so that one is facing Mars at any given time. (Because of the large amount of data that needs to be transmitted to various spacecraft at any given moment, these stations are under constant scheduling crunches and subject to heavy negotiation for available time.) From there, the information was transmitted to the Odyssey orbiter in place around Mars, from which it was beamed down to the Opportunity rover’s flash memory.
Could such early space AI software applications be a precursor of an autonomous space age? Benjamin Bornstein of JPL’s Machine Learning team, who I spoke to at length as we watched the technicians operate on rovers, is optimistic. “We absolutely need people in the loop, but I do see a future where robotic explorers will coordinate and collaborate on science observations,” Bornstein predicts. “For example, the MER dust devil detector, a precursor to AEGIS, acquires a series of Navcam images over minutes or hours and downlinks to Earth only those images that contain dust devils. A future version of the dust devil detector might alert an orbiter to dust storms or other atmospheric events so that the orbiter can schedule additional science observations from above, time and resources permitting. Dust devils and rover-to-orbiter communication are only one example. A smart planetary seismic sensor might alert an orbiting SAR [synthetic aperture radar] instrument, or a novel thermal reading from orbit could be followed up by ground spectrometer readings… Also, for missions to the outer planets, with one-way light time delays, onboard autonomy offers the potential for far greater science return between communication opportunities.”
It’s easy to imagine a future in which automated devices do the work of scientifically analyzing and mining the resources of comets and planets.
Mr. Bornstein also revealed that JPL is developing artificial intelligence technologies for unmanned aerial and aquatic vehicles, and foresees a future in which AI is a regular fixture of the space program: “As the science and exploration goals of future space missions increase in capability and ambition, I believe AI will be one of several enabling technologies. Our approach is to survey current and future missions and ask ourselves what AI technologies dovetail nicely with their goals and requirements.”
Will the descendants of JPL’s Mars AEGIS-upgraded rover be able to reproduce themselves — in other words, will we see a singularity in space? Bornstein doesn’t think that scenario is likely, nor does he see the potential for a 100% autonomous space fleet. He has more concrete goals. Within ten years, he’d be happy just to see artificial intelligence become an accepted part of the space design process. (One exciting possibility he does mention is the potential for multiple spacecraft to collaborate and coordinate with each other, forming a network for team action.)
In the wake of Obama’s redirection of NASA’s moon budget, and in an era where private space companies like Virgin Galactic and Scaled Composites are taking over the human element of space travel, it seems less likely that there will be a substantial government budget for sending human scientists into space, particularly for routine data gathering. Artificial intelligence seems particularly suited for NASA’s scientific recon missions. It’s easy to imagine a future in which automated devices do the work of scientifically analyzing and even mining the resources of near-earth objects, comets and planets. As the Obama administration shifts the space program into bold new directions and forms, it seems likely we will be seeing a lot more AI space technology.