A new event discussing artificial intelligence (AI) applications in space will debut Jan. 18, 2021, space industry executives said in a webinar earlier this month.
The Euroconsult-Innovitech AIxSPACE event, postponed from June 1 this year due to novel coronavirus pandemic-related measures, will be held in Montreal – a city hailed as a major centre for AI research in Canada and globally.
“Artificial intelligence is redefining the frontier of the space industry … and radically transforming the sector,” said Julien Caudroit, Innovitech’s vice-president of communications and marketing, in a “sneak peak” webinar to announce the new date June 1. “Our objective was to bring together key stakeholders in both space and artificial intelligence industries,” he added.
The webinar spoke to changes coming down the pipeline for players in space-based AI applications, from helping astronauts stay safe during a Martian mission, to optimizing satellite data collection that may be different from the new RADARSAT Constellation Mission (RCM) that just celebrated its first anniversary of launch Friday (June 12).
Nathan de Ruiter, managing director of Euroconsult Canada, reminded attendees that AI has long been a staple of space-themed science fiction, from franchises ranging from the lovable robots of “Star Wars”, to the highly influential 1968 movie “2001: A Space Odyssey” that features a wayward AI controlling a spaceship en route to Jupiter. Now, AI is becoming a staple of real life space exploration, including for Canada’s astronaut program.
“We’re currently considering future applications of AI for health care,” said Isabelle Tremblay, the Canadian Space Agency’s director of astronauts, life sciences and medicine. “This will be especially important for deep space missions to Mars. They will last two to three years. The crews will be very remote and there won’t be any possibility to evacuate them,” she said.
Mars is NASA’s next goal for human operations after landing humans on the moon in 2024; since Canada has committed a Canadarm3 (which will have a measure of AI on board) for the lunar Gateway space station, it is possible Canada could join with NASA’s efforts to send humans to Mars in the 2030s or so.
While AI could help crew health on a Martian mission, the results continue to improve Earth-based health care as well, Tremblay said. Already, AI is learning how to parse through large amounts of data to improve diagnoses, such as detecting cancers earlier in their growth cycle. Tremblay foresaw uses for Canada’s northern communities, as well as the quickly growing sector of elder health care. Elder care and northern health are both areas that could benefit from remote diagnoses by doctors assisted with AI, she said.
MDA, best known for the Canadarm series of robot arms and the RADARSAT series of Earth observation satellites, is starting to use AI in its various business streams.
“We’ve transitioned over the decades from robotic operations, where an operator would be at the control of a joystick, to increasingly scripting and aggregating commands into higher-level, more abstract, and more even goal-based commands,” said MDA chief executive Mike Greenley of AI’s applications in robotics.
For example, Canadarm2 on the International Space Station is increasingly operating under “supervised autonomy,” he said, where robotic operators monitor what the arm is doing as it makes goal-based decisions. The data is gathered, parsed and improved upon for future robotic operations and also systems, which could include Canadarm3. This learning process will help gather information, he said, for “five or six years from now, with space stations by the Moon.”
Smaller companies are also using AI for the benefit of multiple industries. Waterloo, Ont.-based SkyWatch Space Applications, which aims to make satellite data accessible worldwide, is using an application programming interface (API) to more easily and swiftly aggregate Earth observation satellite data from multiple sources, which has historically been a difficult problem. At times, these observations need to be very timely, such as during recovery following a natural disaster such as a flood or a hurricane.
“We’re making one API that delivers the data in the same format in a simple way, at a very low cost, that can enable this machine-to-machine capability between the collection all the way down to the consumption,” said Joel Cumming, SkyWatch’s chief technology officer. He said the API is helpful for satellite operations as well as for computer vision models.
“Whenever we can move humans out of the loop, that means the data can flow faster from the sensor down to the application,” he explained. For the example of computer vision models, “They can look at things like image quality. They can look at things like classifying pieces of the image, like clouds. You can imagine if a sensor is delivering data to an application, you don’t have time for a human to inspect that to be able to ensure quality.”
Satellite operations is another area that could see some changes, Cumming said, pointing to the RADARSAT Constellation Mission (a joint MDA/Canadian Space Agency project) as an example that could be altered in future generations of satellites. “If we could optimize and say ‘Hey, in fact, there are going to be other captures either by commercial providers or by other open data satellites out there,’ we don’t need to spend the expenses of capturing with RCM where we can capture in other places. There’s a huge cost savings that can come.”
And already, AI in combination with the Internet of Things is leading to a new level of monitoring of Earth infrastructure, he said. “One of our customers who’s doing this is in the oil and gas pipeline monitoring business,” Cumming explained.
“They built an application that helps oil and gas companies monitor their pipelines in Europe, in this circumstance. What they’ve done is they’ve built a change detection model using SentinelOne [security] data – so free, open data – and they’ve divided up a pipeline from their customers into one square kilometer areas. They detect change with every collection cycle of Sentinel One over those areas. When something pops [in the pipeline], then they can fire a request to our API and request a satellite image of that area that’s changed.”
Even companies outside of the space sector are paying attention to the business possibilities for AI. Phil Beaudoin is co-founder and senior vice-president of research at Element AI, a Montreal-based firm looking to help firms “operationalize” AI for applications in government, insurance and other sectors.
“One of our very interesting results is also in the area of Earth observation. Data is easy to get at the low resolution, but it’s hard to get at the high resolution. Sometimes the high-resolution version is where the interesting detail is going to be,” Beaudoin said. “What we’ve managed to do is when you have multiple images of the same area, from multiple scans of observation sensors, we can combine all of these images together and get the higher resolution image. We do that using AI.”
Another common use of AI is machine learning, or helping to train machines for efficiency in parsing information by using training datasets and then testing the machines on real-life datasets. In space exploration, one possible application could be optimizing the manufacturing process. Creating rockets, satellites and other hardware is an expensive proposition under some traditional types of manufacturing, unless there is a large enough supply line to buy commercial off-the-shelf parts. Machine learning could provide another solution, Beaudoin said.
“What we’ve built is a product called Builder Scout that’s really good at helping capture details about the manufacturing process, even if these details are just in natural language form, or it’s just like, here’s an example of a situation that went wrong,” Beaudoin said. The product will capture the details, he said, and “disseminate them … throughout your company and throughout, also some of your partners.”
Greenley said MDA is also finding uses for AI in high-volume manufacturing, such as robotics-based manufacturing and inspections. In the future, he foresaw opportunities for AI increasing in orbit, as networks of Earth observation satellites work in collaboration to become more efficient, and as satellites and ground stations work on “infusing and aggregating and extracting information out of the datasets that we’re collecting.” Further out in the solar system, he added, AI will be necessary for Martian robotic systems to operate on their own, when a round-trip call to Earth can take several dozen minutes to accomplish.
Element AI’s Beaudoin cautioned that although AI shows much promise in space applications, it is very important to make it reliable. “We can build models that work 99 percent of the time, but in space that one percent might mean people die, right?” he pointed out. For these reasons, he said, humans should remain in the loop of computer decision systems and not be replaced completely. He also called for creating “explainable systems” to help humans better interpret how the models are making their predictions, to avoid the dreaded “black box” effect where computers make decisions with little human understanding of how they came to a conclusion.
The participants also said there should be frequent stakeholder conversations about AI as the technology evolves, to ensure the legal and operational framework in space exploration keeps up as best as possible with changes in the technology.
Observers of the conference were polled on two questions. The first was where artificial intelligence would make the biggest impact in the future; Earth observation and data analysis was the clear winner with 65 percent of poll participants choosing that option. Space exploration received 20 percent, satellite manufacturing and operations 11 percent and satellite communications 4 percent. (The number of participants was not disclosed.)
Another poll asked, outside of the field of defence and intelligence, where the most growth in Earth observation data and insights sales would be in the next five years. Here there was a wider spread of top contenders, leading with agriculture (19 percent), then infrastructure, energy and utilities (18 percent) and transportation (12 percent). The other options were emergency services (6 percent), natural resources (5 percent), finance (4 percent) and communications (4 percent).