攻克行业级柔性操作难题!招商局狮子山人工智能实验室首次亮相WRC 2026
TL;DR - China Merchants Group’s Lion Rock AI Lab debuted its LiOS edge-cloud infrastructure and autonomous dual-arm clothing-folding system at WRC 2026. The system targets the sim-to-real and systems-integration barriers that make deformable-object manipulation difficult to deploy on real robots.
- LiOS unifies distributed model training, simulation evaluation, robot runtime control, remote intervention, and real-world data collection into a closed-loop infrastructure.
- The lab reports over 5× higher training throughput and over 4× faster simulation evaluation, supported by an in-house vision-language-action model.
- Its cloud video pipeline reportedly delivers camera data to GPU memory in roughly 30 ms, enabling cloud inference while preserving real-time robot execution.
- Demonstrations covered multiple dual-arm platforms, autonomous folding of shirts and trousers, and recovery of tangled or heavily wrinkled garments; the system also won the ICRA 2026 LeHome Challenge.