Not to mention that the design constraints of a reaction wheel and a drive wheel are almost polar opposites: a reaction wheel needs to have the largest possible moment of inertia, to maximize the change in angular momentum for a given mass and torque; meanwhile, a drive wheel should have the smallest possible moment of inertia, to maximize the change in angular speed for a given mass and torque, particularly since you typically want to minimize unsprung mass (in fairness, the demoed robot didn't have any suspension to worry about).

Big off-road tires are in the 100 lbs weight range, though. It might work for a special forces dune buggy. https://en.wikipedia.org/wiki/Light_Strike_Vehicle

They might even want that actuated tail, to have a camera/IR sensor at the end of..

Having ridden a lot of motocross in my younger stupider years I can attest to the fact that controlling the rotation of your wheels while airborne is a very important consideration when jumping a motorcycle. If you know what you are doing you can use the rotation or braking of your wheels to control your attitude (and subsequent landing) in the air.

At least from RC car to small motorcycle it seems to scale, no idea about the physics in flipping a humvee and whether the wheels have enough leverage to do that.

I can see the markers for a motion capture system in the video.

Measuring or accurately calculating the orientation at a rate high enough for this is a challenge in itself.