Mind-Blowing Humanoid Robot Walked Outside (The Internet Exploded)

Engine AI built three humanoid robots: the SE1 industrial model, the SA1 research version at $5,400, and the PM1 compact bot that hits 2 m/s. A viral video of the SE1 walking outside sparked the internet reaction, and the company showed them off at CES 2025 using deep reinforcement learning and neural motion control.
Chapters
There's been a viral video going around showing engine AI's SE1 humanoid robot casually walking outside the company's Shenzhen headquarters. And people can't stop talking about it. Some viewers thought the footage was fake CGI because the robot's movements looked too natural. But Engine AI insists it's real. They haven't officially said whether a human was controlling it remotely or or whether it was moving on its own, but they have highlighted that their advanced algorithms and joint modules are what give it that smooth human like gait. It's all happening at the same time. Engine AI made its big debut at CES 2025 in Las Vegas where they showed off three humanoid robots, SE1, SA1 and PM1, each aimed at different markets. The SE1 is the company's first general purpose full size humanoid. It stands at about 1.7 meters tall, weighs 55 kilograms and was designed to handle industrial tasks like heavy lifting and Precise assembly. Engine AI's team says it uses a blend of deep reinforcement learning, imitation learning and harmonic force control joints, plus an end to end neural network model to let it move in ways that look more like what a human would do. They say the se1 can can squat, do push ups and even run, which is pretty unusual for robots of this kind. Engine AI has been trying to tackle issues that have long plagued humanoid projects stiff or awkward movements, and they're proud of how fluid SE1 is in daily scenarios. Engine AAI's presence at CES showed the world the SA1 as well, which is a 40 kilogram bipedal humanoid built for research and education. It has an open source platform so schools and labs can customize the software or hardware to fit whatever projects they're working on. Its aluminum alloy exoskeleton makes it strong and impact resistant and they say it can run and jump if needed. The company says it only consumes around 200W while walking, which is fairly efficient for a robot that size. They priced it at US$5,400, which is a figure that drew some attention because it's lower than a lot of other humanoids in this space. Engine AI mentioned that the initial orders for the SA1 exceeding their expectations, so they had to ramp up production to meet demand. Then there's the PM one, which is a newer humanoid designed to be lightweight, agile and fully open for developers who want to push the limits of what humanoid robots can do. It's about 1.38 meters tall, weighs around 40 kilograms and has 24 degrees of freedom. One eye catching claim is that it can spin its waist 320 degrees, which helps it pull off some pretty acrobatic moves. The PM one is supposed to offer both a mechanical gait mode and a more natural human gait mode, and a lot of it comes down to the software running on its onboard computing suite. Engine AI mentions compatibility with x86 architecture and Nvidia Jetson Orin modules so it can run cross pencil algorithms and handle things like real time object detection and advanced motion control. They're marketing the PM1 to both commercial and educational buyers, saying it's a nice fit for interactive demos, research or even customer service situations. There's a promotional offer that lasts until the end of March 2025, where both the commercial and educational editions are set at around US$13,700, or about 88,000 yuan, with a chance to upgrade from one version to the other at no extra cost. Some statements say it's closer to US$12,000, possibly due to exchange rate differences or different configurations, but the ballpark is in that low teens range. Engine AI is led by founder and CEO Zhao Tong Yang, who has been in the robotics game for a while. He launched Dogotics in 2016, focusing on humanoid robots, then co founded Xpeng Robotics, where he helped develop a quadruped robot that became successful with worldwide. After that, he worked on a humanoid called PX5, which caught attention at Nvidia's GTC 2024. Once that project took off, Zhao left XPENG Robotics and started Engine AI in 2023, raising about 100 million yuan in angel funding, roughly US$13.64 million. They first showcased the SE1 in late October 2024 and then came to CES 2025 with the entire lineup ready to be seen by a global audience. Zhao says the goal is to accelerate the embodied intelligence revolution and push towards an era of artificial general intelligence. He wants robots that can handle real tasks in real environments at scale, not just stand around at trade shows. Engine AI plans to produce more than 1,000 units by the end of 2025, with which is an ambitious goal, but they seem confident in their ability to ramp up production. During their ces presentation, Engine AI explained that the SE1 can do more than just basic walking. They say it's designed for tough factory work, like lifting heavy items or placing parts in precise locations on assembly lines. Because it's built with advanced joint modules, it apparently won't topple over too easily when holding awkward objects or moving across uneven floors. They also claim that all three of their robots benefit from the same general approach to AI motion control. The company used deep reinforcement learning to let the robots gain skill through trial and error, and they combined it with imitation learning that's based on actual human movements. That's how they say they achieved a gait that's more fluid than most older humanoid bots. The SA one is also turning heads in academic circles because it's so open and because it's less expensive than many bipedal research platforms out there. Students or lab members can dig into the robot's firmware, operating systems, or hardware attachments and add new sensors or modules as needed. That kind of openness can help accelerate robotics research because teams don't have to start from scratch. They can build on top of what Engine AI has already developed and possibly share improvements with the broader community. The SA one weighs about 40 kg and can handle a fair amount of wear and tear, which matters in lab settings where things can get banged up pretty quickly. The PM one is different in that it's more compact but still capable of advanced movement. Engine AI notes that it can walk at speeds of up to 2 meters per second, about 4.4 miles per hour, which is faster than some of the robots from other companies, though not as fast as Unitree's H1, which apparently hit 3.3 m s in a recent demo, or Robot Era's Star One. Still, 2 m s is quick enough for many real world applications, but the PM one's 320 degree waist rotation and 24 degrees of freedom let it do certain tasks or with extra flexibility, maybe for creative demonstrations or for fitting into tight spaces. It has a built in interactive core screen that reminds some people of Tony Stark's Iron man interface, giving users an easy way to issue commands, check status, or watch sensor data. The fact that it's also open source under the hood means you can drop in your own AI or robotics frameworks and see what you can make it do. Engine AI's vibe is that the going after that concept of embodied intelligence, where the AI learns by physically interacting with the real world. They want to move beyond pure simulations. They talk about how letting the robot experience real scenarios walking outside or picking up boxes trains the AI in ways that a purely virtual environment can't. This approach can help the robots improve their balance, object handling or pathfinding, and that's especially relevant for industrial environments with random obstacles or unexpected workflow changes. The company is also aware that advanced AI and robotics can raise safety concerns, so they emphasize that they're developing security protocols to protect data and manage Potential hacking threats. They haven't given a ton of details, but they say that reliability and trust are priorities. They also point to power efficiency, the SA one's relatively low energy consumption and is their example of how they want these robots to be as eco friendly as possible. They realize that if humanoid robots become widely adopted, the energy footprint could be pretty large unless companies focus on sustainability from the start. Engine AI's timeline is ambitious. They keep bringing up that they want to hit over a thousand units in production and sales by the end of two, 2025. That's a bold claim, but they've mentioned they already delivered around 100 units in five months at one point, suggesting there's some demand for these systems. China's ministry in charge of industry and it wants to see mass production of humanoids by 2025. And engine AI is one of several companies moving in that direction. Other players like Fourier, Yubike, Unitree and Tesla are also tackling humanoid robots. And everyone has their own angle. Some focus on cost, some focus on high performance hardware, some focus on open source communities. Engine AI seems to balance all three by aiming for reasonable prices, decent hardware and an open development model. A lot of people are curious about how these robots stack up against Tesla's Optimus project. Elon Musk has suggested an eventual cost range of 20,000 to $30,000 for Optimus, which is higher than what engine AI is charging in these early promotions. But many of these projects are still prototypes or early stage units. It's possible prices and capabilities will shift if any of them manage to crack big production volumes, or if they hit unforeseen technical challenges. Engine AI is betting that by unveiling the SE1 first, along with the SA1 and PM1, they can grab market attention and scale up quickly. They also want to push more of the development community to get involved, especially on the SA1 and PM1 platforms. The viral video was a big win for engine AI. People saw the SE1 strolling outside and the footage got attention on social media sites, especially among those who follow robotics or advanced AI. There was that brief debate over whether it was cgi, which the company denies. What matters is that it got people talking about the reality of humanoids that move in a way that doesn't look stiff or mechanical. It also hinted that these robots might soon be able to function in normal urban settings, not just controlled lab floors. That's a key step if we're heading toward a future where humanoid robots might help with deliveries, retail tasks, or even basic housekeeping. People at CES got to see the SE1 do more than just walk. Engine AI set up demonstrations of it, lifting objects, crouching and doing push ups, which suggests it's robust enough for actual work. They said that industrial scenarios are their top priority right now. They see a niche in manufacturing where the robot can operate machines or lift items while coexisting with human workers. They also said the PM one is an example of how they're leaning into commercial interactions. It's smaller than the SE1, can move fairly quickly, has that rotating waist, and supports real time teleoperation or autonomy, depending on how you configure it. Because it has an interactive screen, it's plausible that it could greet customers in a showroom or store, or help run educational demos in museums or science centers. Engine AI's founder thinks we're only scratching the surface of what embodied intelligence can achieve. He sees these humanoids as stepping stones toward future robots that can adapt to just about anything. The open source side means developers can integrate new machine learning models, vision processing routines or sensor arrays for specialized tasks. The hardware, like the harmonic drives and advanced servo actuators, is designed to manage repeated stress and remain stable. The company also hints that they'll expand the lineup with more robots in the PM or PA series, possibly for home or industrial use. Well, first, the big news about the IPL and the PA series. A big focus is on reliability and lifetime support. Engine AI wants to establish partnerships with universities, factories and research labs, providing them not just with the robot itself, but also with updates, technical advice and maybe even community driven improvements. They're showing up on social media platforms like YouTube, LinkedIn, Instagram, Facebook and X, where they share clips of the robots in motion and announcements about software updates or new features. They've even given out an official support email and phone number for inquiries. They're hoping that by connecting directly with developers and potential buyers, they can form a community around these humanoids. Industry experts see humanoid robots maturing with Engine AI, focusing on practical use and aiming for a thousand units by 2025 to refine performance and advance toward AGI. Is everything moving too fast or is this just how progress is meant to happen? Let me know what you think. Thanks for watching and I'll see you in the next one.










