Humanoid robots showcase skills at Ancient Olympia. But they're on a long road to catch up to AI

Key Points

• Humanoid robots demonstrated various skills, such as soccer playing and shadow boxing, during an event at Ancient Olympia, but still face significant challenges in performing household tasks.
• Experts indicate that humanoid robots are approximately 100,000 years behind artificial intelligence in terms of learning from data and executing real-world tasks.
• Successful development of humanoid robots relies on collaboration between researchers, data companies, and manufacturers, which is attracting significant investment.
• Chinese companies are increasingly showcasing their robots in competitive events, while U.S. roboticists tend to rely on polished videos, reflecting a disparity in public demonstration efforts.

ANCIENT OLYMPIA, Greece (AP) — With jerky determination, robots played soccer, wowed children with shadow-boxing skills and shot arrows on Monday at the birthplace of the Olympic Games.

As they shuffled and occasionally froze for a battery change, their creators and futurologists debated the central question of when robots will be ready to tidy closets and wash dishes.

Outer space before house chores

Despite the explosive advance of artificial intelligence in applications like ChatGPT, their physical cousins — robots with human-like appearances and skills — are lagging years behind.

“I really believe that humanoids will first go to space and then to houses … the house is the final frontier,” said Minas Liarokapis, a Greek academic and startup founder who organized the International Humanoid Olympiad.

The four-day event gathered experts and developers at Ancient Olympia in southern Greece where the flame is lit every two years for the modern Summer and Winter Games.

“To enter the house it’ll take more than 10 years. Definitely more,” Liarokapis said. “I’m talking about executing tasks with dexterity, not about selling robots that are cute and are companions.”

Training material lacking

AI is racing ahead thanks to vast amounts of data readily available online. But training material for humanoid robots is scarce. It involves real-world actions that are slower, more expensive and harder to record than digital data like text or images.

By one measure, humanlike robots are roughly 100,000 years behind AI in learning from data, according to an article in the current edition of the journal Science Robotics.

To catch up, author Ken Goldberg, a professor at the University of California, Berkeley, urged makers to move beyond simulations and combine “old-fashioned engineering” with real-world training. That, he argues, would let robots “collect data as they perform useful work, such as driving taxis and sorting packages.”

The race for useful data

Luis Sentis, professor of aerospace engineering and engineering mechanics at The University of Texas at Austin, said that successful robotics requires collaboration between researchers, data companies and major manufacturers to provide scale. Those partnerships, he noted, are already attracting billions of dollars in funding to develop humanoid robots.

“These synergies are happening very, very quickly. So I do see these problems being cracked on a day-to-day basis,” said Sentis, who’s also a co-founder of humanoid maker Apptronik.

Developers at the Greek event brought their own ideas.

Aadeel Akhtar, CEO and founder of advanced prosthetics maker Psyonic, gained international attention after appearing on the U.S. television show “Shark Tank” last year seeking investment for his company’s bionic hand, which offers sensory feedback.

That data, he told The Associated Press on Monday, could accelerate robot development.

“We’ve built our hand for both humans and robots,” he said. “So we’re closing that gap by actually using the hand of the prosthetic on humans and then translating that (data) over to robots.”

Brain cells

Hon Weng Chong, CEO of Cortical Labs, said that the Australian biotech company is developing a so-called biological computer that uses real brain cells grown on a chip. Those cells can learn and respond to information — and potentially teach robots to think and adapt more like humans.

At the Olympiad, organizers hoped to lay a foundation for annual competitions providing an “honest validation of the progress that has been made in humanoid robots,” said Patrick Jarvis, who with Liarokapis is co-founder of robot maker Acumino.

Organizers limited events to what humanoids could reasonably attempt.

“We were trying to get the discus and the javelin, but that’s tough for humanoid robots,” Jarvis said. “We also can’t say whose robot can do a high jump because you’d have to build special legs … and that’s not necessary for most humanoid robots.”

China is keen to display its robots, the U.S. less so

One company even tested whether its machine could manage the shot put, said Thomas Ryden, executive director of MassRobotics, who worked to “get as many humanoid companies there as possible.”

In the end, several U.S. roboticists came to Greece to speak, but few brought robots.

Chinese companies increasingly showcase their machines at public events, such as Beijing’s first Humanoid Robot Games in August, while U.S. rivals mostly stick to polished videos that can mask failures.

There are exceptions. Elon Musk revealed Tesla’s Optimus in 2022: The prototype walked stiffly onstage, turned and waved to a cheering crowd.

Boston Dynamics went further. Ten years after launching its dog-like Spot, the company had them dance in synchrony to a Queen song on “America’s Got Talent.”

One of the five broke down mid-routine, creating a reality-show punchline, but also highlighting their agility and coordination.

“Can I be honest with you? I actually think — I don’t mean this in a cruel way — it was weirdly better that one of them died,” judge Simon Cowell said. “Because it showed how difficult this was.”

___

AP Technology Writer Matt O’Brien reported from Providence, Rhode Island.