At the 2026 Embodied Perception Fusion and Multimodal LLM seminar on 15 September, Wuwen Zhike solution expert Yan Wei argued that embodied AI is bottlenecked by three things: data that is far harder to gather than for autonomous driving, a missing end-to-end toolchain, and the absence of common standards. Wuwen Zhike’s answer is a closed loop of collect world, generate world and simulate world.
On collection, the company runs a task-level and scene-level system that feeds both imitation and reinforcement learning. Its Deqing site in Zhejiang is the largest embodied-intelligence data field in the Yangtze Delta, with more than 500,000 hours accumulated across logistics, home, hotel, industrial, office and retail scenes. Collection mixes on-robot teleoperation with body-free capture, and a six-camera headband device auto-annotates 6D pose and hand pose. Field collection covers factories, stores, homes and offices with vision, force, tactile and inertial data, and body-free data lifted task success by about 20 per cent in basket storage and sorting.
On simulation, Wuwen Zhike built a world simulator to cut the cost and raise the variety of training data. Its Scan2Sim pipeline reconstructs scanned environments to millimetre fidelity, beating common methods by about 5 decibels on PSNR and 0.05 on SSIM, while Gen2Sim generates scenes from large models more cheaply. Natural-language instructions can spawn variations in material, light and shape, and the pipeline exports to USD and URDF while handling tricky objects like cables and soft bodies.
The two flywheels reinforce each other, and Wuwen Zhike joined the MIIT embodied-intelligence standards committee formed in July 2026 to help build benchmarks. The company was founded in May 2023 from Baidu Apollo and Tsinghua talent, entered embodied AI in the second half of 2024, and posted revenue of 25 million yuan in 2024, 50 million in 2025 and more than 100 million yuan by September 2026, doubling each year while serving over 80 per cent of China’s top embodied-AI firms.
Editor’s note: This is an adapted translation of the original Gasgoo report. It has been trimmed and restructured for readability for an international business audience. The full original (in Chinese) is at https://www.gasgoo.com/apps/50640d4b55d5cba175fb84f15d679f19/robot/news/70472293-%E6%97%A0%E9%97%AE%E6%99%BA%E7%A7%91-%E5%85%B7%E8%BA%AB%E6%95%B0%E6%8D%AE%E9%A9%B1%E5%8A%A8%E6%9C%BA%E5%99%A8%E4%BA%BA%E6%99%BA%E8%83%BD%E8%BF%9B%E5%8C%96/.