A humanoid robot ran 100 meters in 9.39 seconds at the World Humanoid Robot Games on Saturday, bettering the human record of 9.58 seconds set by Usain Bolt. Another ran the 400 meters in 38.16 seconds, almost five seconds faster than the human best. Still another jumped almost nine-and-a-half feet into the air — nine-and-a-half-feet! The human high jump record is a hair over eight feet.
But the winner of the 100m race crashed into the padding at the end of the track, reeled backward for maybe 10 meters, then fell and snapped in the middle, sending sparks everywhere.
That image, on top of the impressive speed, the jumping and the other abilities demonstrated at the games in Beijing, strikes me as a pretty good summary of where humanoid robots stand today. Once you sort through all the hype (from Elon Musk, among many others), you see tremendous progress in specialized functions, but you also see that humanoid robots won't show up broadly in factories or homes — or the insurance policies that cover them — for many years.
Let's have a look.
Frankly, I've never understood the need for robots to be humanoid. I mean, if you want to really go fast for 100 meters, you should put a little electric car in the lane on the track. One of those babies will get to 60mph in the first 50 meters — forget Usain Bolt's top speed of a bit under 28mph. The whole race would take maybe three seconds. If you want a robot to set a high jump record, there are these things called rockets.
And specialized robots are already doing great work in a host of settings, as this New York Times article describes in detail:
"Autonomous carts have largely replaced forklifts for moving materials around [a Hyundai] plant. Four-legged robots that look like dogs search for flaws by peering under car bodies. Flat, wheeled robots the size of queen beds slide under Ioniq 5 or Ioniq 9 electric vehicles as they come off the assembly line, lift them from below and transport them to testing stations."
I'd also remind you of the videos we've all seen of the robot forklifts whizzing around Amazon warehouses that grab products from bins so high it would take a person a bit to climb or be lifted there, or that simply pick up bins and take them to people for easy picking and packing. Robots that look nothing like people are being developed that would flip burgers and do other routine kitchen work in fast-food restaurants.
The Times article says auto makers have high hopes for humanoid robots that will complement the massive, one-armed machines they already use to pick up and place car parts and weld them into place, but look at the main use the Times describes:
Hyundai plans to use humanoid robots "initially for arranging components in a specified order for humans to install in vehicles. That is the same basic task that the BMW robot, made by Figure AI, demonstrated in Spartanburg. Mercedes-Benz, which is testing robots made by Apptronik in Austin, Texas, also plans to use them for handling and sorting materials. Carmakers say they have trouble retaining workers to perform such basic tasks."
That sort of use could be very helpful, but it's hardly an example of a versatile robot that could perform an unlimited variety outside of a controlled environment like a factory.
There are two major problems to overcome first, and neither looks to be solvable any time soon, no matter how fast a humanoid robot outpaced Usain Bolt.
The first is that human hands are really, really hard to imitate. Not only are fingertips sensitive in ways that machines have trouble replicating but they manage a combination of flexibility and strength that perplexes robot designers. If they want the strength of the human hand, they need to put a powerful motor in the wrist. If designers want the dexterity, they have to put a host of tiny motors in all the joints where a hand can flex. But if you do both, you wind up with a large, clunky hand. You have to make tradeoffs, and you wind up with a hand that doesn't measure up to the human combination.
The second problem is that, to be as versatile as proponents claim humanoid robots will be, they have to be able to operate in messy environments that robots simply can't navigate at the moment. A factory is one thing, but imagine a robot in your home having to step over the toys left in the family room, not bump into the dog, sort through a whole family's laundry, handling a special item precisely as your teenager demands, and so on. Not gonna happen.
The Wall Street Journal reports that investors are piling into companies working on so-called world models, which use video games and simulations to develop more robust "action" models about how the real world operates, to go beyond the sort of top-down, rule-by-rule, prescriptive models that robots rely on today. But even the enthusiasts say the world models are currently at about the stage where the large language models used for generative AI were with ChatGPT2. That model was released seven years ago.
You'll surely continue to hear lots about the imminent arrival of armies of AI-powered, humanoid robots. They're an enticing idea. Everybody who remembers "The Jetsons" wants their Rosie the maid. There's even a cool term for the trend: "embodied AI." Hypemeister Elon Musk will, alone, make sure that humanoid robots stay in the news, given that delivering 1 million Optimus robots (at maybe $30,000 apiece) is one of the goals he has to meet to qualify for his $1 trillion pay package. And lots of fans
But you can ignore the hype for at least a few more years, both in personal terms and in terms of providing workers comp coverage and insuring homes. Specialized robots will continue to provide great value in industrial settings, but they'll continue on their current trajectory. There won't be some radical change because of a new class of robots. And nobody but the earliest adopters will tripping over a humanoid in their kitchen any time soon.
Cheers,
Paul