Space robotics

Chris Robinson

Space is infamously inhospitable to life, but what is less universally understood is that it is also inhospitable to many technologies. The same high-energy radiation exposure that poses health risks to astronauts from solar particle events also renders the chip in a smartphone, a technology used by billions of people on Earth every day, useless. And while some technologies simply cannot do the job, others are simply not trusted.

This is largely the current situation when it comes to artificial intelligence (AI) applications in space. Space missions are complex endeavours that involve travelling great distances, harsh environments and many unknown factors, and they cost hundreds of millions to billions of dollars from conception to completion. While there is an inherent level of risk that is undertaken to do the work, a certain level of conservatism is also necessary to ensure a mission’s success.

“AI works in a non-linear way. This is the advantage of AI, what makes it so powerful.”

While some labs at York University are creating technologies that are applied to actual space missions today, the work being done to develop AI applications is looking towards a future, exact date unknown, where AI robots will take over much of the labour of space exploration. In some cases, that technology is simultaneously being translated to applications on Earth, where there are far less environmental constraints, but the need for greater trust in AI is no less important, say the researchers at York’s Lassonde School of Engineering who are developing next-generation technologies.

Decades ago, Zheng Hong Zhu’s interest in space robotics was piqued after NASA launched the Hubble telescope, only to announce shortly after that the mirrors Hubble relied on to capture images of distant galaxies were flawed and all the images it captured were blurry. NASA looked into the feasibility of sending robots to fix the issue, but later abandoned this approach, instead sending astronauts. Yet, the idea of using space robots to service and fuel crafts intrigued Zhu and it remains an area of research interest to the present day. Now a professor at York and director of the Space Engineering Design Laboratory, he is developing robotics and AI future applications for MDA Space – Canada.

“AI works in a non-linear way. This is the advantage of AI, what makes it so powerful. But because of its unexplainable nature, because you cannot disentangle it, the space industry does not currently trust it, but is still very much interested in exploring the future of these powerful potentials,” he says.

Currently, explains Zhu, every time a satellite or craft is launched into space, an exact replica is created and left on Earth, so if something goes wrong, engineers can work with the replica to pinpoint the error and fix it. AI, in not showing its work, does not lend itself to such easy corrections.

While colleagues at MDA are attempting to create explainable AI to address this concern, Zhu is working on other pieces of the puzzle, often not involving the most powerful and complex AI technologies, but simpler ones that might be able to be adapted to space sooner. These include simulations to train AI vision for the low-level light conditions that exist in space when approaching a spacecraft; robotics and AI swarming technologies that involve several AI robots that are not the most powerful, but can work together to do more complicated tasks; and training AI robots on learning tasks like grip strength. Zhu is also looking at developing lightweight materials that can be used in space as radiation shields, as the sophisticated AI chips developed by companies like Nvidia cannot currently be used in space.

Associate Professor Michael Bazzocchi, director of the Astronautics and Robotics Laboratory at York, says that while his field used to be focused on fairly traditional methods, that is beginning to change. “Previously, this work has been dominated by classical techniques, but I would say in the last 10 years or so, we’ve seen a huge expansion where there’s interest in how we can apply machine learning, reinforcement learning, deep learning, deep reinforcement learning and computer vision to these fields.”

While space does pose unique challenges in terms of the environment, Bazzocchi says that many of the technologies they develop for space can be applied to the benefit of humans on earth and vice versa.

“While they’re not the same problem, they have many similarities that allow us to apply related techniques.”

One example from his work is an exoskeletal suit designed for firefighters to reduce the amount of effort required while doing strenuous tasks. Part of this lab work requires motion capture to understand a firefighter’s movements, but also requires employing different optimizations and algorithms to understand how the device might reduce or increase muscle activation in ways that might be beneficial.

"If you could send out autonomous robots first, they could perhaps create the right conditions for the humans to follow.” 

In space, the challenge for astronauts is quite the opposite: zero gravity conditions lead to muscle atrophy and eventually bone loss. The same research and principles can create wearables that purposefully create more resistance for astronauts when executing basic tasks.

“It’s not artificial gravity, because it won’t bring them to the floor, but it will make their movements more difficult,” he says. “When they want to do a task, for example, and they have to flex their arm, there’s a motor that’s resisting the motion so that it is not as easy.”

While the possibilities are exciting, Bazzocchi says that in both scenarios, machine learning and AI are not yet trusted.

“When you’re dealing with humans, you want predictability, and very obvious control that’s not going to potentially do something that’s unexpected and lead to injury,” he says. “And the same thing goes for space, when dealing with these multi-million-dollar assets, you want a certain level of predictability and explainability if something goes wrong.”

Still, Bazzocchi thinks it won’t be long until AI plays a bigger role in space.

“There have been some applications of autonomy in space already, such as for time-sensitive operations where there are long time-delays or for doing data processing. So, for example, creating algorithms that evaluate Earth imagery to detect wildfires is very much already in play.”

Zhu says that one day AI technologies might develop to the point where they can pave the way for creating hospitable living conditions for humans in space.

“Elon Musk wants to send humans to Mars. I think in the short term, it’s very difficult because the astronauts would die, the radiation exposure would be too much. But if you could send out autonomous robots first, they could perhaps create the right conditions for the humans to follow,” says Zhu.

Still, how would we know that the robots, working together and autonomously settling Mars, would still act in the interest of the humans that sent them there?

That’s the billion-dollar space robotics question, and according to the researchers, one we don’t currently have an answer for.