The 2026 World Robot Conference has reached a new peak in popularity, but this year's biggest attraction is no longer backflips and fighting shows, but robotic arms moving parts in factories, dexterous hands sorting medicines in pharmacies, and humanoid robots greeting guests in multiple languages in exhibition halls. As "what tasks they can perform" supersedes "what tricks they can show off" as the core agenda, a profound industrial transformation—from hardware showcase to scenario implementation, and from dedicated control to general-purpose brains—is clearly unfolding across the exhibition halls.
Robots Embark on Their "Professional Careers"
At the product application level, a significant shift is underway. High-difficulty stunts such as roundhouse kicks, jumps, and combat demonstrations from previous years are no longer the main attractions. Instead, robots are now moving parts in factories, sorting medicines in pharmacies, and serving as guides in exhibition halls—they have officially started to "go to work."
In industrial manufacturing, robot deployment is transitioning from production line validation to mass deployment. Competition in industrial scenarios has shifted from isolated technical metrics to scaled delivery and scenario adaptation.
At its booth, UBTECH Robotics conducted live demonstrations of its two major industrial humanoid robots, Cruzr Y1 and Cruzr S2, comprehensively showcasing four key scenario solutions: loading and unloading automotive sheet metal parts, loading and unloading automotive machined parts, bin palletizing and depalletizing, and small-parcel sorting for logistics and e-commerce. These solutions cover the entire chain of industrial production, material handling, and warehousing logistics. Reportedly, the Cruzr Y1 has been deployed at Faway VTEC's warehousing facility, capable of continuous, uninterrupted operation for 3.5 hours. In 2025, UBTECH's full-size humanoid robot sales reached 1,079 units, with over 80% deployed in industrial scenarios.
Focusing on "robots that actually get the job done," ZhiShen Technology showcased its full-stack capabilities ranging from core joints and the main robot body to motion control and intelligent decision-making. Its Gangbeng L2 quadruped robot, featuring a peak torque of 95 N·m, a 5.5-hour battery life, IP66 ingress protection, and adaptability to a wide temperature range, is defined as possessing "industrial-grade capabilities to withstand wind, sand, rain, and snow, and operate stably in complex terrains."
Beyond the factory floor, commercial services and human-robot interaction are also experiencing a surge in practical implementations.
In commercial service scenarios, Leju Robotics' Kuafu integrates vision and UWB (Ultra-Wideband) technology to achieve personnel following and dynamic obstacle avoidance. Supporting multi-language and dialect voice interactions combined with anthropomorphic gestures, it completes the entire process of greeting, guiding, and seeing off guests. This solution has been deployed across 23 provinces nationwide.
LimX Dynamics' full-size humanoid robot Oli takes on the role of a "cyber host," capable of communicating in multiple languages including English, Korean, French, and Japanese. Leveraging the WBT whole-body motion foundation model, it can replicate any long-duration actions of motion-capture actors in real time—from twisting waists and shaking shoulders to practicing Baduanjin—while maintaining perfect balance and stability throughout. Meanwhile, UBTECH's newly launched full-size commercial service humanoid robot, Walker C1, integrates guest reception and guiding, cultural and entertainment performances, and crowd interaction, injecting intelligent vitality into diverse spaces such as supermarkets, exhibition halls, and commercial complexes.
In more specialized medical rehabilitation and special operations scenarios, robots are also pushing the boundaries of precision and physical limits. Haier's AI sports exoskeleton robot W2, designed for outdoor sports, is equipped with AI Gait Algorithm 2.0. It can identify terrain and gait at the millisecond level, automatically matching the power-assist mode with a maximum torque of 16 N·m. Its carbon fiber design weighs only 2.4 kg, saving 30% to 50% of physical exertion. Beijing Changmugu Medical showcased the ROPA6 AI full-orthopedic surgical robot platform, where a single machine achieves a "six-in-one" integration of all surgical procedures for hip, knee, unicompartmental knee, spine, trauma, and sports medicine.
PaXini
unveiled the PX-FOOTRIX plantar multi-dimensional tactile sensor. Based on 6D array tactile sensing technology, it achieves full-area 3D array force perception and 6D force/torque sensing on the sole. It can distinguish between soft and hard ground characteristics, analyze terrain changes in real time, and drive the robot to dynamically optimize gait strategies to adapt to complex terrains, enabling humanoid robots to transition from merely "being able to walk" to "understanding how to walk."
Additionally, the Beijing Humanoid Robot Innovation Center made the debut of its lightweight humanoid robot, Tiangong Omni. Standing only 1.35 meters tall and weighing 39 kg, it performed on-site demonstrations of navigating plum blossom poles, climbing stairs, avatar interaction, and office printing. In the future, its capabilities can be expanded to domestic tasks such as household chore assistance, companionship for the elderly and children, and home security.
Evidently, 2026 has emerged as the inaugural year for scenario-based robot deployment. Behind this transformation lies a cognitive upgrade across the industry, shifting from "building prototypes" to "solving real-world problems." Competition in the robotics sector is transitioning from hardware parameter comparisons to the validation of closed-loop scenarios.
Embodied AI Enters an Intelligence Race
Regarding technical roadmaps, the "brains" of embodied AI are gradually converging from the diverse proliferation seen in earlier years. "One brain for multiple applications" and "one brain for multiple morphologies" have become the unified focus across the industry. Previously, the embodied AI sector was long hindered by the pain point of tightly coupled software and hardware: robots of different morphologies often required separately customized control models and algorithm systems, significantly increasing the complexity of data acquisition and model iteration. Currently, numerous manufacturers are exploring how to solve more real-world problems using minimal data, thereby clearing the path for the large-scale popularization of robots.
At the conference, Galbot unveiled its latest small bipedal humanoid robot, ET1, alongside its two veteran "star" robots, G1 and S1. All are equipped with its self-developed general-purpose brain, "AstraBrain." Notably, the ET1 also features the AstraBrain-Agent, a native agent for the physical world, which can acquire entirely new motor skills solely through human interaction, without the need for motion capture or reference videos.
Zibianliang introduced the end-to-end model WALL-A. Leveraging a single foundation model, it requires only short-term, few-shot training to excel in highly diverse tasks within open environments, such as handcrafting sachets, parcel sorting, and home cleaning. Meanwhile, Ant Lingbo launched its Full-Stack Brain 2.0, achieving bidirectional compatibility across different morphologies and brands. According to on-site staff, this brain was already adapted to 17 robot brands and 20 hardware configurations during the pre-training phase. It can execute the entire workflow of autonomous order receiving, medicine sorting, and product delivery in smart pharmacy scenarios, and is equally capable of deploying on industrial production lines for tasks such as grasping, transporting, loading, and picking.
Meanwhile, world models also saw a concentrated debut of physical robot achievements at this year's conference. Daxiao Robotics introduced the Kairos World Model 3.1 and the Environment-Centric Data Collection Solution 2.0, supporting the robot's brain in achieving continuous self-evolution during real-world operations. At the Chaowei Dynamics booth, on-site visitors engaged in intense "human-robot sparring," testing the robot's capabilities in visual perception, trajectory prediction, motion planning, and whole-body control within high-speed dynamic scenarios. This performance is also underpinned by its self-developed KAI world model.
"Throughout the entire industrialization process, embodied AI comprises two components: physical hardware and intelligent capabilities. The value center of gravity will gradually shift from the physical machine to intelligence," stated Gao Jiyang, CEO of Xinghaitu. Unlike most booths that focused on motion demonstrations, Xinghaitu's presentations were entirely centered around "real work," comprehensively showcasing the company's foundational technical accumulation in full-stack software and hardware, as well as brain generalization. Leveraging "brain generalization," multi-level breakthroughs have been achieved. For instance, robots can autonomously schedule multi-step continuous tasks, and in the face of environmental changes, task interruptions, or execution anomalies, they can autonomously correct errors and restore processes, stably completing long-horizon complex operations.
Wang Xingxing, founder of Unitree Robotics, which just went public yesterday, also stated in his keynote speech on the 20th that the explosion of AI technology is significantly accelerating the evolutionary pace of the robotics industry, and the tempo of industrial development will further quicken over the next decade. He revealed that AI models are currently one of the areas receiving the largest financial and human resource investments at Unitree Robotics, and the team is fully committed to implementing a self-evolution system for physical AI robots. Driven by cutting-edge foundation models, this solution allows AI to autonomously retrieve research findings and generate control code. It first validates these in simulation environments before deploying them to physical robots for testing. Ultimately, both AI and humans jointly conduct performance evaluation and provide feedback, forming a complete positive feedback loop that significantly enhances robot development efficiency and skill iteration speed.