This is the equation (no, its not F=ma):
torque*rpm=mass*(acceleration+gravity*sin(angle*PI/180))*velocity*(1/efficiency)/(2*PI)/(#_wheels)
Unfortunately no equation can accurately approximate friction - meaning, it requires measured experiments. Of course, you can approximate it by normal_force*coef_friction, but in reality thats a very rough approximation. And seriously, whats the coefficient of friction between your foam wheels and a persian rug? Or a tile floor? There is no data for this . . . And friction is also dependent on both speed and acceleration.
The only way to know friction is to already have your robot built and tested, which defeats the point of the equation - to design before building.
Sooooo this is why I have efficiency in the equation, to estimate. Friction on the ground only accounts for a percentage of losses, while you also have gear train losses, aerodynamic losses, etc.