?Shaft-hole assembly， as a core manufacturing process， critically determines product precision and reliability through intelligent advancement. Traditional manual algorithms suffer from strong model dependency and insufficient adaptability， whereas data-driven methods exhibit promising generalization capabilities by implicitly learning response patterns from operational data. This paper systematically reviews data-driven shaft-hole assembly technologies， organizing advancements across three dimensions： Environmental perception， assembly control， and task curriculum design. It analyzes agent perception mechanisms， task comprehension-control interactions， and curriculum design impacts in shaft-hole assembly. Addressing current bottlenecks in dynamic response， robustness， and task understanding， the study proposes future research directions： Task-oriented low-cost perception systems， prior knowledge-enhanced data-driven decision frameworks， performance evaluation in unknown task spaces， and task comprehension-control strategies under human-robot collaboration.