EgoSuite — human-behaviour video data
Egocentric human-activity data collected across 7 countries and 25,000+ environment nodes; more than one million hours delivered. Collected once under a standardised system and licensable to multiple customers.
← All companies
Lightwheel 光轮智能
First embodied-data unicorn — world No. 1 delivery volume in three data categories
The first unicorn in embodied-AI data. Lightwheel runs continuous-learning infrastructure spanning human-video data, synthetic simulation data, physical-world reconstruction, model evaluation and real-robot feedback — with the world's largest delivery volumes in human data, synthetic data and simulation evaluation.
Lightwheel provides the data and simulation infrastructure that Physical AI teams train on. Its platform covers real and simulated data generation, reconstruction of the physical world, training and evaluation and feedback from deployed robots, built on a fully in-house 'solve-measure-generate' simulation stack. It is the only Chinese company in both the Newton and EgoVerse international standards bodies, and its simulation assets and synthetic data supply more than 80 percent of leading international embodied-AI research teams.
01Products and solutions
SimFoundry — synthetic simulation data and Sim-Ready assets
High-fidelity, physically credible synthetic data generated at volume from an in-house physics engine and automated physical-measurement pipeline; a standard object becomes a Sim-Ready asset in about 20 minutes.
RoboFinals — industrial-grade evaluation platform
Standardised, reproducible, massively parallel evaluation beyond academic benchmarks, used by the world's leading world-model teams.
HDCP open human-data collection platform and LeIsaac
An open ecosystem with headset, dexterous-hand, haptic-glove, sensor and compute partners targeting billion-scale human data; LeIsaac is listed in the official Hugging Face LeRobot documentation.
Industry continuous-learning infrastructure
Joint ventures and programmes that deploy the data-to-model loop into vertical industries — 'New Light World' with New Hope Group, and Real2Sim/Sim2Real infrastructure for mining and industrial robots with Baotong Technology.
02Key strengths
- Fully in-house simulation platform with an end-to-end loop: a solver layer with its own physics engine (including cable simulation, a long-standing industry hard problem), a measurement layer for high-precision digitisation of real objects, and a generation layer for physically credible synthetic data.
- Rule-maker status: the only Chinese company in both the Newton technical steering committee (with NVIDIA, Google DeepMind, Disney Research and Toyota Research Institute) and the EgoVerse committee (with Meta and Scale AI); co-author of China's national standard on embodied-AI data quality.
- Customers span the top of the Physical AI chain: NVIDIA GR00T, Google DeepMind, ByteDance, Alibaba and Shengshu at the foundation-model layer; Figure AI, 1X Technologies, AgiBot, Galbot and others at the robot layer; Geely, BYD, Li Auto, Toyota, Bosch and Changan in automotive; Moore Threads in domestic compute.
- Infrastructure business model: synthetic data is modelled once and reused indefinitely; human video data is resold to multiple customers at more than ten times — one collection investment, many revenue streams.
- Backed by state and industrial capital — Beijing AI Industry Fund, Zhongguancun Science City, Sichuan Development and Shandong Development alongside Ant Group, New Hope, AUX, 37 Interactive, Baotong and Semir.
- Team combines physics, simulation and industry: founder and chief executive Dr. Xie Chen (physics, Peking University; PhD in quantitative finance, Columbia; formerly head of autonomous-driving simulation at NVIDIA) with core members from NVIDIA, Huawei and NIO.
03Customers and partners
NVIDIA (GR00T)Google DeepMindByteDanceAlibabaFigure AI1X TechnologiesAgiBotGalbotGeelyBYDLi AutoToyotaBoschChanganMoore Threads
Other companies in this category
