At an event marking 20,000 MEgo units and the completion of its million-hour
dataset, Mifeng said it has produced more than 1 mln hours of high-quality,
ontology-free data, all collected in open real-world environments. The dataset
spans 22 major scenario categories, over 10,000 distinct real environments,
50,000+ object classes and 500+ fine-grained tasks, aimed at building scaled
physical-AI data supply.