Thanks for the suggested solutions to the track flipping issue.
Looking at this video http://m.youtube.com/watch?v=fab73SjLymw I think that rather than a simple limit stop you could add torsion bar springs to each axle, to allow fairly free conformation with ordinary terrain and increasing resistance to extreme rotation of the track axle. Not having tested it, my conjecture is that you would go for a spring rate that covers most situations but still ultimately relies on a limit stop rather than spring tension to prevent flipping in extreme cases; i. e. relatively soft, something to add useful tension. Currently, in that and other videos, the lack of tension seems to be reducing efficiency as well as introducing the flipping risk.
I like this idea, I think it would be ideal to keep the spring rate fairly low for positions in the range of +/- 30 degrees so there is little effect on level-ish ground but increase the spring rate for larger angles hopefully keeping the tracks down (pressed against the surface) on steep inclines such as the stairs instead of just preventing them from flipping over. Perhaps this could be achieved with a combination of multiple springs or through a progressively wound spring (not sure of a way to do this with torsion springs tho).
The Mars rover "rocker boogie" design has a cross bar that transfers torque between the left and right side boogies. I don't see such a bar in this design.
The NASA Mars rover uses a differential transmission between the left and right sides of the suspension so that when one side rotates in one direction the other side will rotate in the opposite direction. This robot uses a linkage bar that pivots on the back of the robot connected on each side by ball linkages to achieve a similar motion.
Check out this video to see how this linkage works: Rescue Robot - Tethered Test 1
Is this four indepdenent boogies?
Yes, each of the four track module pivots freely on the suspension. A single track module can be compared to a boogie assembly on a wheeled rocker bogie suspension, the front of the track can be viewed as one wheel and the back as a second wheel with the frame of the track module being the boogie link in the suspension. This makes he entire system comparable to an eight wheeled rocker boggie suspension.
Also, how tight turns can it make without the tracks coming off?
This was one of my concerns with the design but we were able to turn the robot in place without the causing too much of a problem. The tracks did start come off their alignment pins a little after a full rotation but driving forward for a couple of feet got them back in place. During these tests the tracks were not fully tensioned either so correcting that should also improve the performance.