It is done extremely cleverly using CGI, on London's Oxford Street and then recording the reactions for online viral video.
The link to Pepsi is that it's "unbelievable" (like the stunt geddit?) but that's tenuous at best. More likely someone had a good idea and is trying to make it work.
But these things work great largely because they create talkability and moreover, generate a video which can then be shared (already nearly 2m times). Like I just have.