Gasgoo Munich- At the recently opened 2026 World Artificial Intelligence Conference (WAIC 2026), Lightwheel took the wraps off its Physical AI data and evaluation infrastructure. The centerpiece: a continuous learning system built on data, evaluation, deployment, and feedback.
Lightwheel's continuous learning system comprises four components—SimFoundry, EgoSuite, RoboFinals, and RoboStack—which together form a Real2Sim2Real continuous learning loop.
Image Source: Lightwheel
EgoSuite (Data): Delivers real-world experience. Focusing on human behavior data, it accumulates high-quality, scalable, cross-embodiment operational experience. This captures real-world observations, operations, corrections, and long-horizon task execution, providing robots with scalable learning material.
RoboFinals (Evaluation): Serves as an "exam room" for testing robot capabilities. Designed for industrial-scale evaluation, it uses standardized tasks, reproducible environments, and comparable metrics to assess what the robot model has mastered, its capability boundaries, and failure modes—while also defining data requirements for the next round.
RoboStack (Deployment): Channels deployment feedback back into the system. As robots enter industrial sites like factories, warehouses, farms, and logistics hubs, they continuously encounter new task distributions, anomalies, failure samples, and on-site constraints. These insights are looped back into the data, simulation, and evaluation systems, serving as the starting point for the next cycle of learning.
SimFoundry: The physical AI simulation infrastructure underpinning the "data-evaluation-deployment" loop. It tackles a fundamental challenge: how to convert the real world at scale into a simulation environment where robots can effectively learn, train, and validate.
Lightwheel has already achieved global delivery leadership in three key areas: human video data, synthetic simulation data, and industrial simulation evaluation. Notably, the world's top five "world model" teams are all partners. Its client base spans mainstream domestic and international embodied AI teams, robot OEMs, and industrial customers. High-quality data assets are being reused across clients and tasks, with some premium scenario data resold more than 10 times.
This means Lightwheel is no longer offering isolated capabilities for a single link in the chain. Instead, it provides a general-purpose infrastructure capable of serving diverse model paths, robot forms, and industrial scenarios.
