I watched the Boston Dynamics Atlas video in 2013 where it walked across rocks and got pushed by a guy with a hockey stick without falling over. It was cool. It also felt like a science project — impressive but disconnected from anything that would show up in daily life.
Thirteen years later, multiple companies are building humanoid robots that can see, hear, talk, pick up objects, and navigate real environments. The progress in the last two years has been faster than the previous ten, and a lot of it comes down to AI getting good enough to give these machines something resembling judgment.
Boston Dynamics retired its hydraulic Atlas in April 2024 and replaced it with an all-electric version. The new Atlas can rotate its joints in ways a human cannot, pick up heavy objects, and move with a smoothness that the old version lacked. Hyundai, which owns Boston Dynamics, has been positioning the company toward commercial deployment in automotive manufacturing.
Tesla's Optimus has had a rocky public perception. Early demos were mocked — the first reveal in 2022 was literally a person in a robot suit. But by late 2024, Optimus could walk unassisted, sort objects by color, and do simple tasks in Tesla's own factories. Elon Musk has claimed Tesla will sell Optimus for around $20,000 to $30,000, though no shipping date has been confirmed and those numbers should be treated skeptically.
Figure is the startup that got the most attention in 2024 when it partnered with OpenAI to give its Figure 01 robot the ability to hold natural conversations. In a demo, a person asked the robot what it could see on a table, and it described the objects and handed the person an apple when asked. The robot uses OpenAI's vision and language models to understand what it is looking at and what is being asked of it.
NVIDIA entered the space with GR00T, a foundation model designed specifically for humanoid robots. The idea is to give robot makers a general-purpose AI "brain" that understands language, vision, and physical interaction, similar to how GPT-4 is a general-purpose language model. NVIDIA also built Omniverse, a simulation platform where robots can train in virtual environments before being deployed in real ones.
The honest answer: warehouse and factory tasks. Picking things up, putting them down, sorting items, carrying boxes, and doing repetitive physical work. That is where the money is right now, and it is where the technology is most reliable.
Home robots are a different story. Your home is unpredictable. There are pets, kids, furniture in odd places, things on the floor, stairs, doors that stick. A factory is controlled. A home is chaos. The AI that works in a warehouse does not automatically work in a living room, because the living room has a thousand edge cases the factory does not.
The robots that do work in homes right now are specialized: Roomba vacuums, lawn mowing robots, pool cleaners. They do one thing and they do it in a relatively controlled way. A general-purpose home robot that can do laundry, cook, and clean is probably a decade away, if not more. Anyone telling you otherwise is trying to sell you something.
The mechanical engineering of robot bodies has been pretty good for a while. Making a robot that can walk, balance, and grip things is hard but it has been done. The bottleneck has always been the brain.
What changed recently is that large AI models got good at understanding the physical world. Computer vision models can now identify objects, estimate distances, and predict what will happen if you push something. Language models let robots understand spoken commands instead of requiring programming for every task. Reinforcement learning lets robots practice tasks millions of times in simulation before trying them in the real world.
This combination — see the world, understand language, learn from practice — is what people mean by "physical AI." The robot body is the least interesting part. The intelligence running it is what determines whether it is useful or just an expensive statue.
I would not be writing this on a scam-prevention site if there were not a scam angle. There are a few:
Investment scams. Robotics is hot, and scammers know it. Fake investment opportunities in "the next Tesla of robotics" are already circulating. I have seen ads on social media promising 500% returns on robotics stocks or presale tokens for robot companies that do not exist. If someone sends you a link to invest in a robotics company you have never heard of, check the link first.
Fake presales. Crowdfunding scams for home robots have been around for years. Someone posts a slick video of a robot doing impressive things, takes presale money, and either delivers a product that barely works or disappears entirely. The AI-generated demo videos make this easier now because you can fake a convincing product video without having a real product.
Deepfake demos. Some robot demos on social media are not real. They are CGI or heavily edited footage presented as live demonstrations. If a company is only showing you carefully produced videos and never lets independent reviewers test the product, be skeptical. Real robots get reviewed by engineers and journalists who can verify what they are seeing.
If you get a message about an amazing robotics investment or a too-good-to-be-true robot product, run it through our message checker before engaging.
The next two to three years will probably look like this: more robots in warehouses and factories, doing increasingly complex tasks. Amazon, which already has over 750,000 robots in its fulfillment centers, will add more capable ones. Automotive plants will adopt humanoid robots for tasks that are currently done by humans on assembly lines.
Home robots will improve incrementally. Better robot vacuums, maybe robot lawn mowers that actually work well, possibly a robot that can fetch things from another room. But the "robot butler" that science fiction promised is not arriving soon.
The most interesting near-term application might be in healthcare and elder care. A robot that can help an elderly person get out of a chair, bring them medication, or call for help if they fall would be genuinely useful. Several companies are working on this, and it is an area where the technology does not need to be perfect — it just needs to be reliable enough for a few specific tasks.
I am cautiously interested in all of this. The technology is real. The timeline is longer than the hype suggests. And as with any new technology, the scams will arrive before the products do.
New technology brings new scams. Use our free tools to verify before you invest or buy.