[Bongle]

Voted none of the above, for the "bad habits" reason. A sim would have to be nearly perfect in order to teach you optimal strategies for the real game. You could come away learning lessons, but those lessons could be corrupted by small omissions on the part of the programmers.

Also, I'm not sure there is a robust enough physics and simulation system available. For 2006, you would need to model soft balls very accurately, as the deform-ability of the poof balls were a key part of many team's pickup systems. For 2007 and 2008, you'd need to very accurately model inflatable objects. You'd have to model proprietary electronics and sensors such as the CMUCam and gear tooth sensors, since most teams run in teleop with some level of automation and forcing them to push vastly different sequences of commands than they would in competition would remove much of the benefits of practicing.

I wish I could re-vote, I'd also vote "driver training", in that it could be useful in getting a driver used to operating a robot from a fixed position, but not for practicing the finer points of robot operation.