AMD's World Labs Deal Validates 4D World Models; Chinese Rival Lands $28M Factory Order
机器之心 · wechat · 2026-10-02
Following AMD's acquisition of Fei-Fei Li's World Labs (Li becoming AMD EVP & Chief Scientist), 4D world models are drawing industry attention. Chinese startup Yingshen Intelligence, founded in 2024 by Tsinghua PhD Min Wei (ex-Alibaba robotics lead), has been on a native-4D path two years before Atlas.
- Approach: since the physical world is 4D, LLMs and video models are lossy low-dimensional projections that drop spatial structure, occlusion, contact, and deformation; 4D Gaussian primitives preserve motion trajectories and deformation, aiding robot localization, planning, and control.
- Data engine: the "Yingshen 360" rig generates 4D data in real time from 4-6 consumer cameras on a single consumer GPU (industry norm was 80 cameras), with a 10,000+ hour 4D dataset and a flywheel fed by robot execution results.
- Business lines: first physical-embodiment deployment is a shoe factory in Jinjin, Fujian—flexible-material deformation favors 4D—with a 200M RMB ($28M) super-order and thousands of robots in production scheduling; digital entertainment (4D livestreaming, gaming) expands data intake.
- Thesis: native 4D models may become the next foundation model class alongside LLMs and video generation, and could give AMD a new compute-ecosystem wedge against Nvidia.
More from Embodied
- Wayve robotaxi navigates busy rainy Tokyo streets as early investor takes a ride — nathanbenaich · 2026-10-02
- Tesla Cybercab registrations hit 158 in Texas, with 52 spotted on roads — JOBhakdi · 2026-10-02
- MC-EgoHands turns raw egocentric video into robot training data with <7mm hand pose — victormustar · 2026-10-02
- VirtualMacOniPad virtualizes full macOS on jailbroken iPads, running Xcode and Final Cut Pro — tom_doerr · 2026-10-02
- HIDE Benchmark Exposes Memory Gaps in Robotic Manipulation Under Partial Observability — Yansong Shi · 2026-10-02
- InterEvolve Evolves Reward Programs at Test Time to Teach Humanoid Robots New Skills — UIUC-CS · 2026-10-02