October 30 News: Zhiyuan, in collaboration with Tsinghua University and Infinigence AI, has launched the "Collective Experience Learning" (CEL) paradigm. Built on the FERT failure-guided experience retrieval algorithm and the FATS fault-tolerant asynchronous training system, it enables "robots teaching robots" through a 96-unit real-machine cluster, reducing reliance on human expert demonstrations. In tests across five real-world tasks—including flower arrangement, clothes folding, and mobile phone packaging—the CEL solution comprehensively outperformed traditional training methods, with its overall success rate up to 16.7 percentage points higher than the average level of human teleoperators. Under the same training duration, the larger the cluster scale, the higher the robot operation reliability—breaking the industry bottleneck of "insufficient supply of human demonstrations" that hinders the large-scale deployment of embodied intelligence.