HTTP does not use filesystem abstraction. The path in HTTP requests is completely unrelated to files and just looks like a file path.
FTP has actual stateful filesystem traversal commands. PWD / CWD / CDUP / ... . Of course it can be faked, but why bother?
I understood devices in the comment as personal devices for end users rather than servers. In that case discovery is really not external. Same with NAT - it's a difference between "kind of works sometimes" and "is designed to work with it". See Skype's initial success. Or how SIP documents NAT behaviour extensively https://tools.ietf.org/html/rfc6314
The advantage that FTP has is ubiquitous and broad device support. FTP can provide the transport and authentication for a huge variety of devices and provide an instant user story of how to connect with older devices. Discovery and NAT traversal are external to the transport protocol anyway and would be just supporting _ftp._tcp records and the normal TURN/STUN servers.