Use friction between cylinders to simulate gears and chain/belt?

[brettle]

Has anyone already investigated using friction between rotating cylinders to simulate gears and chain/belt? The idea is similar to the gears demo but instead of actually using collision between teeth, represent the gears by cylinders and have them overlaps to create a constant contact joint. This seems to address what I expect are the primary issues with colliding teeth: 1) the time required to detect collisions (cylinders can be collided much more quickly), and 2) the fact that if the driving gear spins too fast it can effectively skip teeth.

Here is a proof of concept of a 90-degree bevel:

friction_gears.zip

To do chain/belt, add an invisible large cylinder between the cylinders representing the two belt pulleys or chain sprockets.

To allow arbitrary (even out-of-plane) transmission of rotational motion between 2 points, put an appropriate bevel connection on each end and connect the bevels to an intermediate rotating rod.

To simulate backlash, use the existing HingeJointWithBacklash proto.

Wrap the whole thing in a proto to simplify usage.

Ideally, the various cylinders used to achieve this effect would not only be invisible, they'd also only collide with each other. One way to achieve that would be with a physics plugin that does nothing for collisions where exactly one of the two geoms is one of the cylinders. I wonder whether the same effect could be achieved with appropriate ContactProperties and a special material for the cylinders.

It might also be worth investigating using spheres instead of cylinders for faster collision detection and possibly simpler math.

Anyway, has anyone already pursued this? This would presumably also address @tlemarec's desires for something like the JointLink idea that was abandoned in #1365.

[ad-daniel]

Well, ODE offers a Transmission joint. There have been attempts to create a corresponding node in webots, the alluring aspect of it is that it has backlash modeling embedded in it, meaning potentially it could significantly faster than HingeJointWithBacklash (which models it using cascaded joints and intermediary solids, making it extremely computationally expensive). One issue that was observed with it was that over time it wasn't very reliable and important errors accumulated, and I think it had to do with the fact that the joint used the CFM/ERP of the world which in certain applications where we want those to be high/low, it made the joint misbehave. That could be just a bug on ODE/our side though, other joints can use a local CFM/ERP, so there shouldn't be a technical constraint for the transmission to do the same (on paper). I don't remember all the details, the idea ended up being abandoned in favor of modeling the backlash using that PROTO (for this and other reasons) but perhaps there's ways around it and/or there have been improvements on the ODE side since.

I'm not saying there's currently a need to support all types of transmissions offered by ODE, but if memory serves right the "Chain-drive" one worked pretty well, and I can see it being useful when modeling something like a SCARA robot.

I believe the groundwork of what you describe is already present in that joint. The other transmission modes were also tested, namely the "Intersecting-axes" mode in order to model angled bevels, but generally the difficulty with those was making it user friendly to model with it, rather than "problems" per-se, it wasn't very intuitive.

[ad-daniel]

PS: if you compile ODE, there's a very nice demo of it here: https://bitbucket.org/odedevs/ode/src/master/ode/demo/demo_transmission.cpp

[omichel]

I believe this friction_gears demo is very nice and it would be great to add it in the samples/devices demos alongside the hinge_with_backlash and coupled_motors demos. I would however remove the teeth from the gears as I believe they are misleading, and use simple cylinders instead. This demo deserves a different use case than the hinge with backlash and the coupled motors. Thus it would be a great addition there.