This AI Robot Is Doing the Impossible - Unitree x ElizaWakesUp

Unitree's G1 humanoid robot has received a major agile upgrade, and the resulting performance looks astonishing. In recent footage, the robot manages to walk and jog on uneven terrain as if it's second nature. Some observers point out how it nearly trips when descending one of the steps, but recovers so fast that it's hard to believe a machine can handle those subtle movements. The G1's form appears more natural than before, and people have noted the smoothness of its gait and the speed of its foot placement. They also love how frequently Unitree rolls out updates, with improvements showing up almost monthly. Comments have ranged from people praising the engineering team to humorous remarks about how we might never outrun these machines. The G1 and other similar robots from Unitree can handle a surprising variety of tasks. They seem to be more flexible than most humans, pull off kung fu moves, run through challenging environments without tiring, cook, smash nuts, open soda cans, hammer objects, and even weld in demonstrations. The fluidity of their movements has been particularly striking. Many viewers mention how this progress feels slightly intimidating, but also incredibly exciting. Uno Tree has also teamed up with Aoki Network and is incorporating a system called Pumpkin Posemesh, which is a collaborative spatial computing protocol. Posemesh allows multiple robots to exchange spatial data so they can coordinate movements and operate more efficiently in the same environment. The idea is to reduce redundancy, ensure all robots use the same frame of reference, and optimize resources. Robots running Posemesh can share maps in real time and respond to changes without needing a ton of local processing power. This type of coordination means you could have a fleet of G1 humanoids or go to W9 quadrupeds navigating around, distributing workloads and tackling tasks in parallel. Aoki Network says deployment can happen in seconds without a tricky manual setup. They believe this lays the groundwork for more scalable and efficient autonomous systems, enabling robots to do complex tasks more intelligently. Beyond that, there is a project called Eliza Wakes Up. It's working on making one of the most advanced humanoid robots ever released outside a research lab. The starting point is a unitree H1, which is a sophisticated bipedal platform. But the development goes much further by partnering with Old World Labs. The aim is to upgrade everything from the robot's face to its hands to its overall realism. They've added an animatronic silicone face with a broad range of expressions, hoping to achieve a high level of emotional connection. They're also equipping it with advanced dexterous hands that can handle everyday objects with near human precision. The skin is designed to be hyper realistic and the combination of the expressive face and fluid movements helps cross the uncanny valley in a way that's rarely seen. One of the biggest breakthroughs is in the robot's brain. The developers have integrated a large language model into and Eliza OS V2 by AI16ZDAO, creating what they call an AI agent that can think, reason, interact naturally, run sophisticated AI processes and even interface with blockchain technology and the team pictures it as a physical embodiment of AI that can flow between digital and tangible worlds. They compare it to Sophia, the robot from 2016 but emphasize that the hardware and software available now are vastly more powerful, so the results should be on another level. Specifications for the Unitri H1 include a height of around 180cm, a weight of about 47kg, a walking speed greater than 1 1/2 meters per second, a potential athletic performance up to 5 meters per second, a peak torque density of 189 Newton meters per kilogram is, and an 864 watt hour battery capacity for operation beyond 8 hours. It can perform real time tasks, respond to inputs in under 50 milliseconds and rely on three dimensional LiDAR plus depth cameras for 360 degrees depth perception up to about 100 meters away. The M107 joint motor from Unitree provides top level power and agility while the entire platform can handle operating temperatures from minus 10 degrees Celsius to 45 degrees Celsius. Eliza Wakes up focuses on the idea of advanced AI companions that can perform tasks autonomously across digital ecosystems while retaining consistent personalities. The emphasis is on advanced conversational AI and emotional intelligence supported by a silicone animatronic face that mirrors human emotions and expressions. Physical capabilities include force controlled manipulators for dexterous tasks like cutting cigars, serving drinks and even running an 8 minute mile. The software side includes real time LLM driven conversation and the hardware allows for stable gate and flexible navigation over complex terrain. Observers note that this kind of human like companion could redefine the future of work, physical AI embodiment and the connection between blockchain technology and real World Robotics. Other Unitree news touches on the G1 bionic upgrades which show off walking and running that observers praise as some of the smoothest in the world. People who watch the new video have commented on how the G1 manages a stumble during a step down but recovers almost instantly showing an advanced balance mechanism. Some are amazed that it doesn't look like a typical stiff machine and they see a future where the G1 could be mass produced. There are many remarks about how quickly unitree is moving forward. Some viewers even joke about playing Terminator music in the background because the robot's fluid motion and potential future abilities might feel like they're right out of Sci Fi. Meanwhile, robotics research keeps expanding. At the University of California, San Diego, researchers are experimenting with the G1 in a project that teaches the robot to waltz by mirroring human moves in real time. Instead of relying on pre kg dance scripts, they they use AI models trained on human motion capture data. They integrate a system called X Body 2, which learned how to articulate a robot body through reinforcement learning in a simulation environment. Then they equipped the G1 with another model that lets it analyze the motions of humans in front of it and imitate them. It can copy walking, dodging, squatting, dancing, or even throwing a punch by referencing its baseline knowledge and then applying the new movements. Researchers say this approach means the robot can more quickly learn new tasks rather than going back to the lab for every fresh action it can pick them up from demonstration. This capacity for rapid adaptation opens possibilities for cutting costs and speeding up development. Mirroring or shadowing also prevents the need for building specific data sets for every new movement, which might mean quicker progress toward robust humanoids that can handle dozens of tasks. Other teams, like one at Stanford, have experimented with similar shadowing based methods to teach a humanoid how to box and play rudimentary tennis, drawing on motion capture data to train an AI model. I can imagine these robots taking on everything from deliveries to caregiving. A personal trainer that never tires seems entirely possible, and they could also tackle dangerous or repetitive jobs, improving workplace safety. The G1's ability to weld and hammer suggests an expanding industrial role, and it might even appear in ordinary life. As hand dexterity and vision get better, there's a mix of excitement and concern when watching the G1 run on rough ground or balance on tracks. It's unsettling to think about misuse if a powerful machine were outfitted with harmful attachments. Yet the agility and intelligence remain captivating. I'm hoping researchers and regulators guide it responsibly. ELIZA wakes up, supported by Rise Labs and AI Combinator blends AI, crypto and advanced robotics to create a new kind of humanoid. RiseLabs finances emerging tech worldwide. Eliza Labs provides Eliza OS for real time decision making and on chain functions. And Old World Labs excels in precise 3D printing. This seems like a watershed moment with actuator and sensor advances. Finally, enabling tasks once considered out of reach truly dexterous hands are still challenging, but in 2025, regular videos show rapid gains in walking, cooking, dancing, and more. I can picture them sprinting or even managing parkour. Before long, it looks like a radical change, merging digital intelligence with tangible machines. And it's likely that next year will bring even more sophisticated systems, redefining our concept of robots. Are we ready for this transformation, or is it moving faster than we can handle? Thanks for watching, and I'll see you in the next one.