Electrodacus, I have proven your assertion wrong, and proven a vehicle constructed as I described, works.
I constructed a simple vehicle with a thread spool and a worm gear on a vertical axis, and a pinion gear and driving wheels on a horizontal axis. I also created a
LeoCAD model of the vehicle, so that anyone with Windows, Linux, or Mac, can examine the vehicle in 3D, and build one or a similar one with whatever parts they may have handy. Just save the attached trike.txt as trike.ldr, and it will open in LeoCAD. It only uses parts that are available in LeoCAD by default, I believe.
When the spool turns clockwise, the single wheel is the rear wheel.
When the spool turns counterclockwise, the single wheel is the front wheel.
Using strong, non-springy polyester thread (that I normally use for sewing buttons back on), configured either way, pulling the thread off from the spool makes the vehicle move; even when you are pulling the thread in the opposite direction.
Specifically, if the two wheels point left and one wheel right, and the thread comes off the spool counterclockwise and through a hole in the black holder, pulling the thread right makes the vehicle travel left.
Conversely, if the thread comes off the spool clockwise and through a hole in the gray holder, pulling the thread left makes the vehicle travel right. In other words, this works –– so that pulling the thread makes the vehicle travel in the other direction –– in both directions and configurations; all you need to do is swap the direction the spool turns when you pull the thread off it.
Because this is a worm gear drive, there is absolutely no energy storage. When you stop pulling the thread, the vehicle stops, because the wheels cannot turn the spool (as a pinion cannot drive a worm, only the worm can drive the pinion). The spool axis has minimal inertia, too; when you stop pulling the thread, the spool does not unwind on its own.
I can take video of it tomorrow if you insist –– I need to make some kind of a holder for my phone and get better lighting, because my hand-held video was horrible ––, but I'd prefer you yourself build and test a similar vehicle. It does not need to be exact same, just make the gearing ratio small enough so that there is sufficient stiction for the wheels to drive the vehicle forward. Note that I had to use a worm gear because I didn't have pinions of different sizes in this set, and because the driving wheels are so much larger than the spool, I'd have needed a large reduction anyway to get the surface speed ratio below 1:1.
I seem to have misplaced my Lego tubs, but I did have one
unopened set, Lego Technic 9395 "Pick-up tow truck", that I got as a present from a friend. This limited the types of gears I had available. As the set has no chain elements, the best I could do was a thread spool and a single worm-pinion reduction. I would have used spur gear reduction gearbox if I had suitable gears; alas, this set does not have sufficient gears to do that. But it does have the worm and pinion.
To turn this into the original treadmill model I described, I would need a Lego chain, two sprockets for it, and replace the spool with a sprocket at the bottom of the trike. Two idlers for the chain on the trike, so that the chain makes an Ω-shaped loop around the sprocket, would help ensure the chain does not slip. If you don't want to run the chain vertically, you need to replace the worm and pinion with spur gear reduction instead. Remember, the reduction has to get *surface speed ratio* below 1:1 for stable running to be possible, but that does *not* limit the ratio of how fast the vehicle travels compared to the thread or treadmill speed (see \$v_c\$ and \$v_t\$ in my earlier post, and how they relate via \$\lambda\$, the surface speed ratio.)
The smaller the ratio, the easier it is to get movement without losing traction. Remember, *any* movement of the driving wheels moves the vehicle away from the direction you're pulling the thread or the treadmill surface travels, so something like a 1:10 reduction just makes sure you're not tripping on insufficient traction/stiction and such issues.
I do not think anyone should trust my word for it if they truly doubt this is possible (and nothing out of the ordinary, just a trivial mechanism).
This is why I show the CAD model of the exact vehicle I built and verified works as I described –– and Electrodacus claims is impossible ––: so that you can examine it and build it or your own version of it, and prove it yourself. There is absolutely no trickery here.