China’s New 98% Accurate Robot Just Got SCARILY Capable!

[00:00:02]China's humanoid robots are improving quickly, and two new updates show just how far they have come. Jaomi's Cyber 1 has reached a 98% success rate after spending an entire quarter inside the company's electric vehicle factory. And with its redesigned body, the latest version of Cyber 1 looks surprisingly similar to Figure03. So, let's see how the two robots actually compare.

[00:00:26]Meanwhile, Limx has just revealed a new robot operating system that gives its Ali humanoid the ability to handle longer and more complex tasks. Lim X demonstrated this by showing Ali organize an entire messy room through one continuous sequence of actions. Hey guys, Alfie here. Welcome back to AI Nexus. Jaomi's humanoid robots have now completed their first major factory assignment, and the results show how quickly the company is moving from experimental hardware toward real production work. After spending an entire quarter inside Jaomi's electric vehicle factory, the robots reached a 98% success rate at a self-tapping nut station. That is only one percentage point behind experienced human workers, which is why Jaomi is describing the trial as a completed internship. The original task required the robot to repeatedly collect nuts and place them accurately on both sides of the production station. Earlier in the trial, the success rate was around 90.2%.

[00:01:24]Jaomi then improved the robot's sensing, control, and movement until it could reportedly operate continuously for several hours while maintaining the higher 98% result. But tightening or placing a small rigid component is only one type of factory work. Jaomi has now moved the robots into two more difficult stations where the objects do not always behave in a predictable way. The first new assignment involves sorting side covers used around a vehicle's center console. These panels are large, irregularly shaped, and slightly flexible. Unlike a solid metal component, the panel can bend, shift, or become caught while the robot is trying to move it. Siomi says the robots are already completing this task with a success rate of around 90%. During the demonstration, the robot reaches deep inside a material box to collect one of the panels. When the object is too far away, the robot places one hand against the edge of the box to support its body while reaching forward with the other.

[00:02:23]This is where full body control becomes important. The robot cannot treat the arm as a separate machine because leaning forward changes its balance. Its legs, waist, torso, and arms must adjust together. Once the panel is lifted, the robot sometimes changes its grip using both hands. This dual arm coordination allows the machine to reposition the flexible object before placing it inside the storage rack. The robot also uses proprioception, which is its internal sense of where its fingers and joints are positioned. That allows it to make small corrections without depending entirely on its cameras. If the panel becomes caught during placement, the robot does not immediately stop or repeat the exact same motion. Shyomi developed an active compliance system that detects resistance and adjusts the angle of the object. The robot can move the panel slightly, test the opening again, and continue until the part slides into place. Shyomi says this is its first longrunning factory demonstration involving flexible automotive components. The second new assignment involves folding and recycling empty material boxes. The robot must first use its fingertips to release a small pull ring. It then folds the container through synchronized two arm movements. Xyaomi has also demonstrated multiple humanoids operating around the same station, showing how several robots could eventually share work across a production area. These new abilities connect directly to Shyomi's earlier redesign of the Cyber 1 robotic hand.

[00:03:49]The previous hand update reduced the mechanism to roughly human hand proportions, added tactile sensing across the fingers and palm, and increased the number of active movements available to the fingers. That design was meant to help the robot detect pressure, slipping, and contact even when its camera could not clearly see the object. Siame also developed a liquid cooling system inspired by sweating to remove heat from the hand during long periods of operation. The company said the redesigned hand survived more than 150,000 grasping cycles during a continuous endurance test. Those upgrades become more meaningful inside the factory because the robot is now using fingertip control, touch feedback, and in-hand adjustments during real production tasks. Human supervision is still part of the system. Trained operators can remotely take control if a robot reaches an unsafe position or enters a failure that it cannot recover from. The robots also still need some materials, including the folding boxes, to be placed in a suitable starting direction.

[00:04:48]So, Shyomi has not yet created a humanoid that can enter any station and immediately replace a skilled worker.

[00:04:55]But the progress is moving beyond a single programmed task. The larger goal is a robot that can transfer the same hands, sensing, balance, and control system into several factory jobs without requiring completely different hardware for every station. But while Shyomi is expanding one robot across multiple factory jobs, Limax is testing the same adaptability in a far less controlled environment. The home Limax is continuing to improve its humanoid robot Ali with a major new software update called Kosa 0.5. In its latest demonstration, the company showed Ali entering a messy room and completing a full cleanup task in one continuous run.

[00:05:37]Instead of performing one simple action, the robot handled laundry, toys, boxes, furniture, trash, and even a final handover to a person. The demonstration lasted nearly 3 minutes, and Limck says it was recorded without cuts or resets.

[00:05:52]Oi first walked into the room and began placing clothing inside a laundry hamper. This may look simple, but clothing is difficult for robots because it does not keep one fixed shape. The robot had to identify the fabric, reach for it, adjust its grip, and carry it without dropping it. Kosa 0.5 uses cameras on Ali's head and wrists to understand what the robot is looking at while the task is happening. The robot then moved toward a pile of objects on the floor and picked up a shark plush toy. Ali bent deeply, grabbed the toy, stood back up, and placed it on a chair.

[00:06:28]This part showed one of the main improvements in Kosa 0.5. Walking, bending, reaching, and balancing are now handled as one connected movement instead of separate actions. When Ali bends forward, almost every part of its body must adjust to stop the robot from falling. After that, Ali lifted a box from the floor and stacked it on top of another box. This required more precise placement because the box needed to land in the correct position. The robot also collected several small toys from a desk and placed them inside a container one at a time. These tasks showed that Kosa 0.5 can handle both larger whole body movements and smaller hand movements within the same system. Behind these actions is Limx's new vision language action system called V3-0. The system has one part that understands the instruction and decides what should happen next. Another faster part watches the live camera feed and keeps adjusting the robot's movement. This means Ali does not simply follow one fixed plan from beginning to end. It can respond if its body stops in the wrong position or if an object is not exactly where it expected. Ali also noticed a chair that was out of position, straightened it, and pushed it toward the table. The robot later collected a used bottle and other trash before placing them inside a bin. Moving furniture is especially challenging because the object is larger than the robot's hand and requires the entire body to apply controlled movement. The final task involved taking a wallet from a shelf and handing it directly to the owner. A handover requires the robot to understand where the person is, approach safely, and place the object where the person can take it. Limx is using this to show that Ali is being developed for environments where robots may work directly around people. Kosa 0.5 also helps Ali recover from mistakes. During testing, a human expert could take control when the robot missed a grasp or became stuck in an ineffective movement. The expert corrected the action and then returned control to Ali. Those mistakes and corrections were later used to improve the robot through realworld reinforcement learning. In one basket cleanup test, the success rate increased from just over 18% to 74% after this training. Across three tested tasks, the overall failure rate fell from nearly 40% to just over 11%. KOSA 0.5 does not mean Ali can already clean every unfamiliar home without help, but the demonstration shows how LIMX is combining vision, decision-m, balance, walking, and manipulation into one connected humanoid system built for longer realworld tasks. In 2026, buying a human-like robot for your home is no longer just an idea from the future. Six realistic robots that you can actually buy right now, and one of them costs only $10,000. Moya has warm artificial skin and a biomimetic body built to move almost like a real person. Ava can detect where you touch it, how much pressure you use, and even the temperature of that contact. UBX U1 can recognize emotional changes and start conversations when someone enters the room. Real Boatexa takes things even further with replaceable faces and a personality that owners can completely change. We have reached the middle of 2026 and Realistic Robotics is moving faster than almost anyone expected.

[00:09:56]Robots that once existed only as prototypes are now available for pre-order or direct purchase. Let's start with Moya. A robot that takes realism one step further by offering something most robots still cannot.

[00:10:10]Warmth you can actually feel. Moya comes from the Shanghai company Droidup, which is also known as the robot is being developed for hospitality, elder care, and companionship. Founder Lee Chingdu describes the idea as a warm robot that feels more comfortable to stand beside and touch. That idea is built into the body. Moya's silicone skin is heated to between 32 and 36° C. That is close to normal human skin temperature. Under the surface is a Walker 3 frame with tendon assisted actuators and lattis style artificial muscles. The face has 25° of freedom, which gives Moya more control over expressions during conversations.

[00:10:52]Moya can also walk at about 0.83 m/s.

[00:10:56]Droidup says the gate reaches 92% humanlikeness. This gives Moya an advantage in hotels and care spaces where a robot may need to move between people instead of remaining in one place. The design also creates problems.

[00:11:10]Moya's walk can still look unnatural and the soft skin may be expensive to repair. The robot costs about $173,000.

[00:11:20]Moya focuses on physical warmth. While Ubek's U1 tries to make the interaction itself feel more human, UB built its name with Walker robots made for factories. On June 30th in Shenzhen, the company introduced UWorld as a new consumer brand. The U1 became its first full-size ultrabionic robot built for homes care and emotional companionship.

[00:11:43]UB is mainly targeting people who live alone and older adults who may want regular interaction inside the home. The U1 has 88 degrees of freedom, which allows its face, neck, arms, and upper body to move in a more natural way. Its dual pivot neck can tilt and turn like a human neck. The silicone skin helps the robot look less mechanical during close interaction. Its emotionaware AI can identify more than 20 emotional states with over 90% accuracy. The robot can then adjust its voice and response based on what it sees. The most unusual feature is that the U1 does not need a wake word. It can notice when someone enters the room and begin a conversation on its own. Most of the processing happens locally, which also gives users more control over private conversations.

[00:12:30]The limits are still serious. The battery lasts only 2 to 4 hours. The robot is designed for flat indoor floors and cannot climb stairs or handle normal household chores. Pricing starts at about $17,600 for the light version. The Pro costs around $25,000.

[00:12:49]Ultra pricing ranges from about $130,000 to $146,000.

[00:12:55]UB Tech reported over 13.3,000 orders during the launch. U1 tries to make the whole body feel alive. A headform takes the opposite approach and puts most of its effort into one area, the face. A headform 2.0 Zero is a fantasy inspired bust robot made for social interaction and entertainment. Founder Hu Yu Han wants to bring digital characters into the physical world. That is why Schwan is designed more like a living character than a general purpose humanoid. Up to 42 small actuators move the face beneath soft bionic skin. Cameras placed inside the eyes help Schwan follow people and make direct eye contact. In February 2026, the robot gained attention after performing a lip-s synced song at an AI gala. The performance showed how much detail the face can create through small movements around the eyes and mouth.

[00:13:48]Jwan works best in live events, exhibitions, and entertainment spaces.

[00:13:53]However, the robot cannot walk and usually remains on a fixed stand or wheeled base. Each unit also takes weeks to build. Pricing depends on the level of customization and can range from tens of thousands of dollars to several hundred thousand. Jan is designed to be watched. Eva is designed for much closer interaction because its skin can detect exactly how you touch it. Eva comes from IO technology under the Robonova.ai brand. Founder Mark Woo presents the robot as a physical answer to digital loneliness. The company describes it as a new kind of relationship between humans and embodied AI. Its main feature is FPC bionic skin that can sense pressure, temperature, and the exact point of contact. Graphine heating warms the surface to around 37° C. The cameras are placed inside removable glasses instead of the eyes. This helps the face avoid the fixed stare seen in many androids. Emotional data is processed offline and stays inside the robot. Ava do.ai is still at an early stage. Its movement is limited to sitting and reclining positions, so it cannot walk around the home. The current official pre-order price is $10,000 with delivery planned for November 2026. Earlier Kickstarter backers were able to reserve units for as little as $6,800.

[00:15:17]The first four robots focus heavily on presence and emotion. Hobs W1 takes a more practical route by giving businesses a robot that can greet people and handle simple objects. Hobs W1 comes from Noetics and is designed for hotels as well as retail stores. It can also work in showrooms and corporate reception areas. The robot has hands with six degrees of freedom and arms with 5 degrees of freedom. It can navigate indoors and show information on a chest display. It can also read basic emotional cues from guests. These features allow HOBS W1 to welcome visitors and point toward different areas. It can also hand over light objects during simple service tasks. The robot is not trying to look fully human.

[00:16:00]Instead, the design focuses on useful interaction at a lower cost. Noix has already shown that it can build affordable humanoids through the Boommy robot, which was priced at about $1,380.

[00:16:13]The company says Hobbs W1 is also being developed around consumer electronics level pricing. However, the official price has not been announced. Its light frame also means it can only handle small items and basic tasks. Hobbs W1 shows how a humanoid can be useful without looking human. Real Bodex Arya takes a different approach by allowing owners to change both the face and the AI running inside the robot. Arya is the flagship social robot from Real Bodex.

[00:16:41]The company has deep experience with realistic silicone sculpting and that background is clear in Arya's face. The robot is also available as a digital character through Ask Arya, which can communicate in 25 languages. The physical robot uses cameras inside the eyes and a magnetically attached face.

[00:16:59]Owners can remove one face and replace it with another. Arya can also connect to outside AI models such as GPT, Gemini, Claude, and Llama. This means the software can change without replacing the body. Real Bodics is already testing Arya in real settings.

[00:17:15]The company has announced pilots with Ericson, a school district in New York, and an elder care program in the United Kingdom. These trials are meant to test how the robot performs in education, customer interaction, and care environments. Arya still moves on a wheeled base and cannot walk on two legs. Independent testers have also reported slow replies and uneven emotion tracking. Prices begin at $20,000 for the bust. A halfbody version costs about $95,000.

[00:17:47]The full wheeled robot starts at $125,000.

[00:17:51]These are six realistic robots you can actually buy in 2026.