Canadian business and government leaders are calling for more investment in artificial intelligence projects to prepare our country for the next revolution in space exploration.
Machine learning (ML) – a subset of artificial intelligence (AI) – allows computers to train from initial human-curated datasets to make decisions on their own, whether that be selecting interesting targets on a space telescope or making a repair on the forthcoming Gateway space station – which MDA’s developing Canadarm3 is expected to do later in the 2020s.
“Our teams like to joke about the fact: imagine leaving your kids at home for a week with a bunch of tasks, and let them just go and do all their tasks, and don’t even check in for like five days, and then come back and see how they did, ” MDA’s chief executive officer, Mike Greenley, said in a pre-recorded message broadcast at the AIxSPACE online conference Monday (Jan. 18).
“That’s what it will be like operating on Lunar Gateway for our teams, in terms of creating an autonomous control environment where we can give the robotics its tasks. Then [we’ll] come back in four or five days and check in see how that went, and give the next series of instructions for operations.”
The keynote session from Greenley and a live panel discussion immediately following explored the applications for ML, and its challenges. Repeatable results is one thing the community wants, to mitigate the old “black box” problem of a computer making decisions with poor explanation for the outside observer. The community is calling on more collaborative projects with other industries and other government departments outside of the usual few to use ML in space.
The Canadian Space Agency is at the forefront of such efforts, with the Lunar Exploration Acceleration Project (LEAP) providing opportunities for companies to receive funding for numerous efforts, including ML. LEAP is in support of the Artemis Moon program, where autonomous operations will be a must to support astronauts working for long periods on the moon’s surface.
During the panel discussion, entities such as Mission Control Space Services and Maya HTT paid tribute to the opportunities LEAP afforded them.
The program, Mission Control’s president Ewan Reid said, “has been a really good thing for us,” allowing the company to work with NASA and international partners on projects such as an autonomous rover. At Maya HTT, ML and data science expertise grew in the past decade with the assistance of LEAP and other funding from CSA, said Remi Duquette, vice-president of industrial AI and data center clarity.
A Western University representative also thanked the CSA for its support in helping fund a new optical device – a LIDAR with a multispectral camera.
“We’ll put some AI into that,” said Kenneth McIsaac, associate (acting) director of Western’s Center for Planetary Science and Exploration. Using AI solutions in scientific instruments, he added, would allow the data to come back to Earth somewhat “curated”, because “boring or bad data is rejected at the source, and only interesting samples get returned … to occupy the time of human scientists.”
Yet challenges remain for the community. CSA’s Erick Dupuis, director of space exploration development, said “space should be a much more important user of AI than right now” given the forthcoming complexity of projects such as Canadarm3. That isn’t to say that the community lacks knowledge in – or desire to work with – AI and ML. Rather, there is more that can be done – and the key actors are working hard to catch up to the terrestrial side of ML and AI, in which MDA is implementing data analytics (as a single example of terrestrial uses of these technologies).
The CSA is working with the National Research Council of Canada’s Industrial Research Assistance Program (IRAP) to “foster spinoffs” from space into terrestrial applications, Dupuis said. Such spinoffs are a common benefit of Canadian space exploration; for example, Canadarm technology has been repurposed into surgical robotic applications that remove tumours.
Already robotics operations are pivoting to the route of automation, if we recall that the Dextre system aboard the International Space Station is now operated by ground controllers in concert with computer software, for example. But as we move out in the solar system, Dupuis said, astronauts and robots alike will face communications delay and bandwidth concerns when communicating back to Earth. The ideal space system, he said, would make decisions in “real time” without the need of a human constantly monitoring it.
CSA is thinking ahead to a new challenge even beyond the moon, which would be implementing a synthetic aperture radar (SAR) system at Mars, building on long-standing expertise in this technology in the Canadian space community (with MDA’s Radarsat satellite series as just one prominent example.)
“Techniques in artificial intelligence would be really interesting to, in this case, prioritize the data we want to send back to Earth,” Dupuis said. Already we are seeing the planetary rovers Curiosity autonomously make small decisions about driving or deploying the laser instrument, and as ML becomes more adept, Dupuis said that such vehicles could make more decisions “without the benefit of a local operator.”
Mission Control’s Reid added that landmark projects such as Canadarm3 will test our ability as humans to “trust the applications we’re developing.” ML thrives in testing computers on known datasets, while space is all about seeking the novel and the unknown, he pointed out.
Moreover, Maya’s Duquette said, space systems are (rightly) built to a high standard and are not prone to failure, which makes it difficult to predict how ML and AI would behave in failure modes. “We’ve worked so hard to get to those places where there are no failures,” he pointed out.