By John Keilman

Humanoid robots marched through Chicago's convention center last week, delivering snacks, shaking hands and busting out dance moves.

But recent viral mishaps elsewhere — a humanoid dancing uncontrollably at a restaurant, another kicking a small child during a performance in China — underscore a big challenge for robot makers aiming to put them to work in factories and warehouses: How can they ensure a humanoid doesn't hurt a human?

Makers of humanoids, which have humanlike attributes to perform jobs a person might do, say they're unaware of anyone who has been seriously hurt or killed by one of the robots. But the machines are getting bigger and heavier, approaching 200 pounds in some cases. At that size, people in the industry worry about the damage that could ensue if a bipedal robot loses power.

"If you do that with a humanoid, it can fall over and crush you," said Michele Silva of the functional safety engineering firm Reynolds & Moore.

De-risking humanoid workers may be key to the sector's aggressive growth goals. Right now, humanoid makers are raising big money. Agility, an Oregon-based company whose robots are already laboring within a Plexiglas cage at an auto-parts factory, recently announced plans to go public at a valuation of $2.5 billion.

While the machines are just starting to enter workplaces such as factories and warehouses, some companies plan to eventually deploy them in people's homes. Morgan Stanley researchers project that a billion humanoids will be in place around the world by 2050, with the total market worth $7.5 trillion.

Traditional industrial robots such as welders, palletizers or loaders are "deterministic," meaning they adhere to a fixed set of rules and produce a constant result. Humanoids tasked with performing multiple jobs use artificial intelligence and are "probabilistic," meaning they operate on statistical likelihoods, not certainty.

That will require the robots to have layers of safeguards before they can work shoulder to shoulder with human beings. Companies that gathered at the Automate conference in Chicago said many of those are already taking shape, beginning with emergency stop buttons and going all the way down to the microchip level.

Chip designer Nvidia announced a new safety system for humanoids underpinned by its cutting-edge Blackwell chips. Amit Goel, the company's senior director of robotics and edge-AI ecosystem, said the system can interpret sensor data about potential hazards, stopping the robot when conditions aren't safe and allowing it to keep working when they are.

"The safety brain and the functional brain need to talk to each other often and with much higher context," he said. "We built that operating-system layer and the software stack so that you can run these two things together."

Aside from their own sensors and cameras, humanoids can be governed by similar devices mounted in the workspace around them. Philadelphia-based Fort Robotics makes controllers and software that process information from multiple sources to let humanoids take nuanced actions, said Samuel Reeves, the company's chief executive.

"It's more complex data about, 'I've recognized a person, they're in this location, in this pose,' and then making sure that the robot has that information and can trust it to a level where they can make safety decisions," he said.

The possible consequences of a humanoid robot losing stability have led an expert panel at the International Organization for Standardization, which creates safety guidelines, to examine the subject. It expects to publish a standard by mid-2028.

Humanoid makers, meanwhile, are coming up with their own solutions.

Neura Robotics, a German company, makes a two-legged, 176-pound model it calls 4NE1. Founder David Reger said it has been engineered to minimize the risk to people nearby. If the robot detects a problem, such as a knee joint that stops responding, it will try to recover its balance; if it can't, he said, it will collapse onto itself like an imploding building.

Some humanoid makers have nullified the issue by eschewing legs. Dexmate, based in the Bay Area, is developing robots that travel on wheeled bases and have long arms that can reach items on warehouse racks. Co-founder Yuzhe Qin said the battery and electronics go into the base, which gives it a low center of mass.

"This kind of ratio makes it super stable," he said. "It will not fall."

California-based Noble Machines contends that putting a robot instead of a person into a hazardous environment can be the greatest safety feature of all. It's developing a two-legged humanoid — the model it brought to the conference wore sneakers — that is designed for construction sites, factories, mines and other rugged surroundings.

"Operating close to toxic chemicals, operating inches away from heavy machinery, is very dangerous for people," said co-founder and chief executive Wei Ding.

Brad Porter, founder and chief executive of Cobot, said the risks from humanoids need to be kept in perspective. His company is developing wheeled robots with arms that can push carts in hospitals or sort parts in a factory. They move at walking speed and don't possess shattering grip strength.

"We don't need to impart a lot of energy into this space," he said. "We're not trying to crush watermelons or anything."

Write to John Keilman at john.keilman@wsj.com