As AI moves into the physical world, is Canada missing the boat on robotics?

A global race is underway to bring robotics into our everyday lives, with a new generation of AI-powered robots promising greater flexibility.

Rapid advancements in artificial intelligence mean robots are being adopted for tasks ranging from working alongside humans in warehouses, to delivering packages on city streets, to inspecting dangerous locations. 

What's more, robots are increasingly capable of learning on the job — and experts say Canada stands to miss out if we don't embrace adoption at this critical time. 

If there's a buzzword in artificial intelligence right now, it's "physical AI" — something that was on full display at this month's Consumer Electronics Show (CES).

The promise is that physical systems, when kitted out with sensors — machines like robots, autonomous vehicles or industrial equipment — can act logically and responsively in the world when paired with current approaches to AI.

At CES, Google and American robotics company Boston Dynamics announced they're teaming up to test AI-powered robots in Hyundai auto factories — two models of a machine named Atlas. 

While experts say we're still a long way from the kind of general-purpose humanoid robots that might one day live in our homes, washing the dishes and folding the laundry, we are at a moment where AI is shifting into the physical world. 

Traditionally, robots are programmed top-down to take on a specific series of steps; that's fine for tightly controlled environments with repetitive, infrequently changing tasks, like the robotics found on a factory floor. 

But using the approach that's led to so much success in generative AI means you can train robots in a bottom-up way, making them more "plug and play," or essentially able to learn on the job.

This opens up robotics adoption to smaller companies that "don't want to have to do coding and a lot of programming," said Hallie Siegel, CEO of the Canadian Robotics Council. 

"When there's sufficient intelligence baked into that process, the robot itself can learn how to complete a task. It doesn't need to be coded."

This newer approach means robots can not only adapt more quickly, but also take on "much more sophisticated tasks," where "you can bring them to a level of reasoning and thinking," said Raquel Urtasun, a computer science professor at University of Toronto and CEO and founder of autonomous trucking company, Waabi.

And machines that need to move safely in dynamic environments, like autonomous vehicles, can be trained in virtual environments. 

At Waabi, Urtasun said, "what we did was build the metaverse for self-driving, meaning a simulator that is as realistic as the real world."