WRC 2026: Chinese Robotics Firms Show Embodied AI Moving from Demos to Real Production Lines

At the UBTech booth, industrial humanoid robots Cruzr S2 and Cruzr Y1 worked continuously from 9 a.m. to 5 p.m., handling car parts loading and unloading, palletizing, and sorting. The robots achieved positioning accuracy under 1 mm and an average picking pace of nearly 1,100 items per hour in a logistics sorting task. Jiao Jichao, vice president of UBTech and head of its embodied intelligence institute, said that what customers buy is the real scenario, not benchmark performance. UBTech's "embodied brain" consists of a three-layer system: Thinker base model for perception, Thinker-WM world model for prediction, and Thinker-VLA for execution. After edge-side optimization, inference efficiency increased by 176% and storage usage dropped by 60%. The company says real-robot data accounts for 60-70% of training data. UBTech reported 2025 revenue exceeding 2 billion yuan and total humanoid sales of 13,838 units, with over 80% of full-size humanoid robots deployed in demanding industrial scenarios.

Magic Atom transformed its booth into a "magic town" replicating industrial manufacturing, logistics sorting, and public safety scenarios. Its wheeled humanoid MagicBot D1 performed material handling and transfer tasks, while a heterogeneous multi-robot system with XGZ1 and HyperBot demonstrated a complete supply-to-sorting workflow. The company also launched a light industrial quadruped robot, MagicDog T1, and showcased its Magic-VLA K02 general-purpose embodied model performing long-horizon tasks such as block printing, folding clothes, and box sealing, with a success rate over 90% for the latter. Magic Atom said it has orders worth over 1.1 billion yuan, products covering nearly 30 countries, and overseas revenue accounting for 40%.

Galileo Robotics, a national-level "little giant" enterprise, unveiled Galileo X, a new ground mobile platform that integrates the capabilities of AGVs, off-road vehicles, and quadruped robots into a single unit. The company said it developed a "unified ground embodied mobile system" that breaks down the barriers between wheeled, vehicular, and legged hardware, motion models, and control algorithms. Galileo X is positioned as the core strategic product for its global expansion over the next three to five years.

OriginFlow, a startup founded by Tsinghua doctoral student Qin Shentao, presented NeuroScale, a data collection system based on surface electromyography (sEMG). A wristband captures muscle electrical signals and, combined with first-person vision and IMU data, generates "Human Tokens" that represent human motion and force. Qin said the system aims to provide physical AGI with a "motion foundation" beyond text and video. OriginFlow has raised over 500 million yuan in cumulative financing. The company is at the first stage of a three-phase plan, moving from human data collection to enhanced human capabilities.