Continue reading this on our app for a better experience

Open in App
Floating Button
Home Views Tech

Humanoid robots are here

Assif Shameen
Assif Shameen • 9 min read
Humanoid robots are here
Forget all the high falutin sci-fi stuff from the future. Look around you, the robots are already here / Photo: Ark Invest
Font Resizer
Share to Whatsapp
Share to Facebook
Share to LinkedIn
Scroll to top
Follow us on Facebook and join our Telegram channel for the latest updates.

Readers of this column might recall a piece I wrote about travelling on Waymo robotaxis in Phoenix, Arizona, late last year. More recently, I wrote about Los Angeles-based Serve Robotics, whose robots deliver Shake Shack burgers in Hollywood and bibimbap (a Korean rice dish) in the city’s Koreatown to customers’ doors.

If you are a tourist walking around Hollywood or searching for bulgogi, as I like to do whenever I am there, you are likely to bump into one of the robots every few minutes. They are everywhere, on every sidewalk and street corner, almost like they own the streets of Los Angeles.

Forget all the high falutin sci-fi stuff from the future. Look around you, the robots are already here. They are not just delivering food to movie stars, or driving journalists like me in robotaxis or helping to perform complex robotic surgeries in hospitals, they are also helping nurses tend to the sick and caregivers with the elderly. They are in factories helping to manufacture our everyday products, or welding and painting cars, or in farms picking fruits and vegetables. They are in stores restocking the shelves or at the back of pharmacies sorting pills and medicines. As drones, they are making themselves useful for surveillance or gathering intelligence. They are also in warehouses packing the boxes from Amazon and Costco that I find outside my door almost every other morning. Suffice to say you can no longer ignore or avoid them.

The word “robot” was first used in 1920 by Czech playwright Karel Capek in his play Rossum’s Universal Robots to describe human-looking machines used for mundane labour. Not long ago, I visited one of the highly automated Infinite Kitchens that fast-casual restaurant chain Sweetgreen runs in New York. A friend had recommended I buy the fast-growing firm’s shares so I decided to check out its kitchens. Their robots prepare salads in bowls, allowing human staff to focus on the order’s finishing touches. Infinite Kitchens can produce 400 to 500 bowls an hour, or 50% more than a regular restaurant, with a far smaller staff. Eventually I decided not to buy Sweetgreen shares, which is a shame because the stock is up 150% since I passed on it.  

In Southeast Asia, smart delivery robots like Pudu Bellabot, that cost just US$15,000 (around $20,000) a piece, are now increasingly common in KFC, Pizza Hut and the Filipino Jolibee franchises delivering food and beverages and even interacting with guests. During my last two trips to Japan, I recall seeing quite a few variations of sushi robots.

Here is the thing: Pudu Bellabots, sushi robots, the tray robots in restaurants in China and Japan, or my own floor-cleaning Roomba, or logistics and delivery robots don’t look anything like a humanoid robot. Amazon uses a robot that has an arm, a camera and monitor on its head. The narrative until recently was that while automation was indeed becoming ubiquitous with robots in factories, warehouses and more recently food preparation and delivery, the “real robots” or the ones we humans are afraid of were still way out in the future. No robot was ready to take our jobs away. Why should anyone be afraid of a taxi that drives itself? Maybe some taxi drivers might lose their jobs to a bunch of fancy robotaxis in a few cities, but the rest of the billions of us will still have our jobs, including 99% of taxi drivers and Grab drivers — or so the storyline went. The humanoid robots were at least a decade or so away.

See also: Nvidia CEO says he was surprised that publicly-held quantum firms exist

The demand for robots

The world really needs a ton of robots.  For one thing, the demographics just don’t look good. There just aren’t enough humans to do the jobs. 

Japan is a rapidly ageing society. More than 10% of Japan’s population now is over the age of 80 and currently 31% of the population is over 65 years of age. China’s demographic problem could turn out to be even worse than Japan over the next few decades. China, whose population peaked at 1.43 billion in 2021, will see its population shrink to just 630 million in 74 years, according to UN’s World’s Population Prospects. That’s a decrease of over 56%. 

See also: Google agrees to buy cloud security firm Wiz for US$32 billion

Someone has to grow food, deliver groceries, prepare food in restaurants and, of course, write the code for AI-powered software to automate the factories which will make robotaxis and flying cars. Fortunately, we will have the robots to do these tasks.

The arrival of robots should boost productivity and transform an array of  industries. Over the years, automation’s impact on productivity has changed the face of industries. The advent of the washing machine reduced the time needed to do the laundry by 87%. After Henry Ford installed an assembly line a century ago to produce Model T, the first mass-market motorised vehicle, automakers needed 88% less time to manufacture a passenger car. The arrival of Kiva robots at Amazon’s warehouse a decade ago cut online shopping time from click to ship by 78%.

What’s next? Think humanoid robots, or AI-powered machines designed to mimic human motion and interaction. What’s so special about the human form factor? The key is that a humanoid robot is generalisable. “While a wrench can tighten nuts better than a human hand can, it is not a generalisable tool,” says a recent report on robots by ARK Invest, a tech-focused US asset management firm. “The human hand is generalisable, particularly in an environment built by and designed for humans.”

In late February, I watched a Youtube video featuring start-up Figure AI’s Helix, its new in-house generalist vision-language-action (VLA) model for humanoid robots, which unifies perception, language understanding, and motion control. Helix uses a single set of neural network weights to learn all behaviours and enables multi-robot collaboration. 

Figure AI launched its first humanoid robot, Figure 01, in early 2023, just weeks after ChatGPT was unveiled. Around the time, Figure AI raised US$70 million in its Series A round from venture capital firms at a US$400 million valuation. Last year, it raised US$675 million at a US$2.6 billion valuation. Last week, Figure AI raised another US$1.5 billion at a whopping US$40 billion valuation. That’s a 100-fold increase in just two years. 

The humanoid opportunity

Humanoid robots can understand and execute tasks just by listening to us, as shown in online videos of these robots performing tasks in the kitchen via simple voice commands. One video shows two of Figure AI’s second-generation humanoid robots, Figure 02, working together to figure out whether an apple, Snickers chocolate bar and a box of tissues need to be placed in the refrigerator or straight into the pantry. 

Sink your teeth into in-depth insights from our contributors, and dive into financial and economic trends

The system is powered by two AI components: a multimodal language model with 7 billion parameters that processes both speech and visual data at 7–9 hertz, acting as the robot’s brain. The second is an 80-million-parameter AI that translates these instructions into precise motor movements at 200 hertz. Helix controls 35 degrees of freedom in real time, allowing for intricate motions from individual finger movements to full-body coordination of the robots. 

The two robots worked to organise food inside a refrigerator without any prior exposure to any of the specific items they were given. The system only required 500 hours of training data, or significantly less than other similar AI projects. Since the entire system runs on embedded Nvidia graphics processing units (GPUs) inside the robots, it is ideal for commercial use. These robots can handle things as fragile as eggs and place them in the refrigerator without making a mess on the floor.

Unlike traditional robots that require specific programming for each task, Helix enables them to interact with new objects and situations without prior training. Figure AI CEO Brett Adcock says that Helix is a major step toward making robots useful in our everyday life. While Helix’s real-world effectiveness still needs to be fully tested, its ability to adapt without constant reprogramming is a game-changer, Adcock says. 

Figure AI partnered with OpenAI to develop Helix, but the two firms have since parted ways. Now, Figure AI is developing its own AI models for high-speed robot control. Meanwhile, OpenAI recently started hiring engineers for a new in-house robotics team.

More than 20 companies from around the world, mostly American or Chinese, are racing to produce their own humanoid robot. Besides the nine humanoid robots shown in the photo, Nvidia, Apple, Meta Platform, Open AI or China’s Baidu are also developing their own. So, when will we see the first commercial humanoid robot? Tesla has a 2027 target release date for its Optimus humanoid – just a few years away even if it was delayed. I believe by then, independent players like its creator Figure AI will be formidable competitors for the big tech giants like Amazon, Tesla, Apple and Google.  

So, just how big is the humanoid opportunity? ARK Invest estimates that household robots will eventually be a US$13 trillion opportunity, while the industrial robots market could also be around a similar size by 2030. Tesla CEO Elon Musk believes there could be over one billion humanoids on earth by the 2040s. 

Bank of America, in a recent research report, forecasts sales of one million in 2030 and a robot population of over three billion in 2060. Bank of America estimates the Tesla’s Optimus Gen 2 humanoid robot’s content cost would be around US$50,000 to US$ 60,0000 now, if all of the major components are made outside China. If most Optimus hardware is made in China, the content cost can be lowered to US$35,000 by the end of this year and further decline to US$17,000 per humanoid robot by 2030. 

The sweet spot is probably under US$10,000, possibly closer to US$ 5,000. To get there, you need economies of scale like smartphones or selling hundreds of millions of humanoid robots. People in ageing societies would happily pay US$5,000 for a humanoid robot that would do household chores like helping out in the kitchen and doing their laundry.

Humanoid robots like Kime are already pouring and serving drinks for customers and handing out snacks at self-contained kiosks in Spain. Other humanoid robots are working as hotel concierges or have taken other customer-facing roles. In the education industry, NAO and Pepper, developed by Aldebaran Robotics (now part of United Robotics Group), are working with students, creating content and teaching programming. Pepper is designed for human interaction and social applications, while NAO is more focused on STEM (science, technology, engineering, and mathematics) education. In hospitals, humanoid robots are providing services like communicating patient information and measuring vital signs. The future envisioned by Capek in his play has already arrived.  

Assif Shameen is a technology and business writer based in North America

×
The Edge Singapore
Download The Edge Singapore App
Google playApple store play
Keep updated
Follow our social media
© 2025 The Edge Publishing Pte Ltd. All rights reserved.