Robots in real world will almost never act the same way in simulation.
That's a strange statement.
And also a very accurate statement. I work for the US Navy in robotics, and we use simulation only to verify new code, and when we really don't want to risk damage/losing the million dollar robots. And even then simulation isn't really regarded as a good test of something new.
In the few years I've worked there and the handful of tests I've seen, never has a field test ever matched simulation the first, tenth or any try in between.
Now I may have to preface that with one if: If you are talking about simulating a Roomba bouncing around a square room, I bet a real Roomba bouncing around a square room will match. Simple environments will easily match simulation. But when we talk about multiple autonomous underwater vehicles cooperatively hunting for and classifying targets on the sea floor, simulation doesn't even begin to come close to real world testing. It comes down to how much of a controlled environment are we talking about?
For simulating a robot design, I don't know of any programs. Robots are highly complex, it would be hard to create real world physics models. I mean, you could have noise generated by a servo effect sensor readings, tolerances in components differ, there's a million simple things that could affect a robot that it would be hard to model them all.