Well, you don't have many choices with that type of situation, as you would need some type of parser that would be able to determine that you've indicated coordinates instead of a whole number:
"i am at 75, 55" would normally be interpreted as, 'I am at 7555'. You could get a little bit more verbose with wildcards and say something like, "I'm at 75 dot 55" and use the {TXTPOS} token to find the, 'dot' (whether the speech engine renders that as, 'dot' or '.' is unknown unless I test it) ... Use {TXTSUBSTR} to get what's on either side of, 'dot' and then get the numeric values from each side by using {TXTNUM}. The command would be something like, 'I'm at *'... which (as a wildcard) is relatively unreliable.
You could also use an inline function written in C# or VB.net.
There's also the new {CMDSEGMENT} token, but depending on the range of the coordinates, that could generate a LOT of commands in the background:
i'm at [-100..100][-100..100] would make 40000 commands which would probably work best, but it would totally cripple your experience.
Hope some of that is helpful.