Are humanoid robots on the verge of remaking the economy, or ‘doomed to fail’?

What can a humanoid robot actually do in the real world?

If you’ve seen videos online of the Boston Dynamics Atlas robot or Tesla’s Optimus bot, you might have assumed that these machines are just about ready to deal cards at your Friday night poker game, do your laundry and apply for a job on the assembly line.

A visit to the Living Lab at MIT last week gave me a different impression: bots that stand on two feet and attempt to do simple tasks like stacking bricks or tossing a tennis ball still perform like toddlers, at best. They require patient instruction, and you shouldn’t expect a predictable outcome.

The Living Lab is a new area set up for robot experiments, and its centerpiece is a mocked-up apartment kitchen, complete with fridge, sink, center island and cabinets. It sits inside the Computer Science and Artificial Intelligence Lab on MIT’s campus.

That lab, known as CSAIL, has spawned robot companies like Boston Dynamics and iRobot, the inventor of the Roomba robotic vacuum. It’s regarded as one of the top university labs exploring what’s next in robotics.

“We’re creating an environment that captures real-life properties of the spaces where we want to put machines in,” said Daniela Rus, director of CSAIL. “So that’s a real kitchen with a real fridge and a real dishwasher and a real oven. We can make cookies in the kitchen. … In the past, our robots have been in research environments, where the environment is heavily simplified over the kinds of things we expect to see in the physical world.”

A robot chef whips up some lemonade

In that kitchen, an $80,000 humanoid robot from Rainbow Robotics in South Korea is positioned behind the island, but instead of legs, it has a wheeled base.

A grad student on a laptop gives the robot a command, and it begins making lemonade from a powdered mix. A pitcher has been pre-filled with water, but it can use a spoon to scoop mix from a bowl, drop it into the pitcher and stir. Then it pours the lemonade into a wide-mouthed glass.

While the robot moves somewhat slowly, the demo goes off without a hitch. The robot was trained by having humans perform the same actions while wearing sensors.

It had a slight advantage, though; it was told where the objects were on the table, so it didn’t have to identify or locate them on its own. Jared Boyer, a Ph.D. student at the lab, explains that this demo has been in development for about a year.

Another demo involved training a robot to lay bricks — albeit lightweight foam ones — to create a pyramid shape. A human assistant fed it bricks one at a time, and the robot did a good job of picking them up, but when it came to stacking them, it would occasionally knock a foundational brick out of place. This robot, a Unitree model made in China, also placed the top brick such that it immediately fell. It didn’t yet seem to have a quality control routine built in to detect what had happened and fix the issue.

The last demo we saw was another Unitree bot programmed to toss a tennis ball at high speed. But it didn’t have a hand that would grasp and then release the ball at the right moment. Instead, its right arm ended in a scoop that resembled a jai alai player’s cesta. On the first attempt, the ball left the little scoop so fast that it was impossible to see, ricocheted off a wall, and then disappeared from sight.

I’m not sure who asked for a second toss, but this time the ball went flying, the robot lost its balance, staggered and fell face-first onto the carpeted floor. It needed help getting back onto its feet.

I’ve definitely seen cutting-edge demos go off the rails before, so that wasn’t a surprise. But it did make me feel like an all-purpose workplace or household helper may still be a few years off. Same for a robotic replacement for Yoshinobu Yamamoto, this year’s World Series MVP.

Rus, the lab’s director, acknowledged as much.

“The humanoid robots are very complex mechanisms. They’re very difficult to control,” she said. “It will be a long time until the humanoid robots will do the wide range of tasks that we imagine them to do. To make humanoids better, AI has to get better.”

What does that mean?

Adding more intelligence to the brains of these robots, she said, will help them understand “how to control the body, how to adapt to the task … how to understand and interpret the world around them and how to operate within the world.”

An east-west divide

There is an East Coast/West Coast divide in the robotics field. Many of the most successful Massachusetts robotics companies have built bots that can reliably perform a single task, such as vacuuming a floor, venturing into dangerous situations to assist police and the military or moving merchandise around an Amazon warehouse.

Many of the organizations attracting the most funding in the robotics field right now are West Coast companies working to build extremely capable humanoid robots, like Figure, which has raised $2 billion.

Elon Musk has predicted that humanoid robots — including those made by his car company, Tesla — will remake the global economy in the next 15 years.

But it’s not hard to find a robotics entrepreneur in Massachusetts who believes that trying to develop viable humanoid robots is “doomed to fail,” in the words of Bruce Welty, a founder of several robotics companies that focus on automating warehouses.

“Think about making eggs in the morning,” said Sidney McLaurin, a Boston robotics entrepreneur. What kind of package are the eggs in? Do they smell OK? What happens if a piece of shell gets into the bowl? “This is just one of an infinite number of tasks that are really easy for humans,” McLaurin said.

He said he is “bullish on robots that either do one dull, dirty, dangerous, high-value task really well or augment the ability of a human to enable them to do something faster, more efficiently, etc. But we are a long way off from replacing humans.”

While research like the projects at the Living Lab may lead to more capable machines that can operate in human environments, McLaurin is building a startup by focusing instead on doing a few of those “dull, dirty, dangerous” tasks today. His startup, Fleet Robotics, is designing robots to inspect and clean cargo ship hulls while they’re in transit.

And while it’s more common to find that sort of robotics company in Massachusetts, we do have one significant competitor in the humanoid robot Olympics.

That would be Atlas, the robot that Boston Dynamics of Waltham unveiled in 2013, and has been improving ever since. This year, Atlas is getting its first high-profile job: working in a Georgia factory owned by Hyundai, the South Korean manufacturing firm that also owns Boston Dynamics.