A device with both cellular and Wi-Fi has historically treated them as separate networks with separate addresses and separate sessions. Convergence means treating them as two paths to the same session, and the user plane function is where those paths come together.
Access traffic steering, switching and splitting is the mechanism. Steering chooses which access carries a flow. Switching moves an established flow between accesses without breaking it. Splitting uses both simultaneously for a single flow, with a multipath transport reassembling at the anchor.
That anchor is the user plane function. It terminates both the 3GPP path and the non-3GPP path — the latter reached through an interworking function — and presents one session to the data network. Which is why placement and convergence are the same conversation: the anchor has to be somewhere sensible relative to both accesses.
The practical constraints are unglamorous. Splitting a flow across two paths of different latency requires reordering and buffering. Device support varies. And a Wi-Fi network the operator does not control offers none of the guarantees the scheduler provides on licensed spectrum, so the steering policy has to assume the worse path can degrade without warning.