Pic: 4143 differential swerve concept



This is a thought project to remove the steering motor from a swerve drive module. Mars/Wars used this as an offseason modeling project. The idea is to use two separately controlled wheels per swerve module to both drive and steer.

So each wheel is geared to one motor, and to turn you just turn the wheels opposite directions? Or is there actually a differential in there?

Mechanically elegant - essentially just two nano style gearboxes and an idle thrust bearing.
However, presuming two drive trains and rotation is obtained by spinning the wheels at different speeds, you would probably need three encoders per module. One would be for each wheel and a third to measure rotation, as rotation is likely to drift considerably from the difference in rotation of the wheels.

I love the idea (seriously awesome) but encoders are the least of its problems. I think the motor controllers are going to bed in a constant, losing battle to keep the module pointed straight, as one wheel picks up more traction than the other, and as the whole bot gets pushed around sideways.

That said, you could get away with two encoders, since there’s only two axes of control. One on the motor, and one on the steering spindle. The second motor would just be driven at a percent power of the first. You’d probably want three encoders for redundancy anyway, but even that won’t compensate for the lack of direct steering control imo.

Now if you really wanted to go gung ho with this design, drop one wheel and driver it with an actual differential. The housing could be bolted to the chassis, and the rotation of the module could be set by the relative speed of each motor, with gear teeth, rather than carpet, to push against. It’s still not going to work well, but it’s closer!

(the response for steering is way faster than the response for speed, so unless your drive acceleration is very slow, it will be difficult to keep the steering controller from over shooting.)

Actually having to go into a pushing match doesn’t sound fun with this.
Plus your module has increased MOI; not great from a controls standpoint. (I’ve wondered how big of a deal MOI is on swerves for a while now…)

This is an interesting concept for FRC, and one that has seen use in plenty of industrial systems.

For FRC, your next iteration is to make it …

… triple coaxial :ahh:

For a modelling proof, there are some really nice features (symmetry, use of TB Nano, low custom part count) in this design.

If you didn’t wamt steering motors on the modules, couldn’t you just put a gearbox in the middle and belt-drive it to each module? Saves motors and weight.

Say this was geared for 10ft/s and the two wheels on the module were 2 inches apart on center. Not taking friction into account, and assuming there’s enough torque to turn the module, the module would be capable of rotating at approximately 1200rpm.

This is awesome. Always impressed with what 4143 comes up with!

I’m afraid if the motor allowance rules remain as they were last year, you’d only be able to make 3 of these modules.

If at FIRST you want to succeed, try triangle robot?

I really like this concept. Maybe it’s not the most practical thing in the world, but not “wasting” any motor power on steering and being able to just have a big ol’ dead axle tube to pivot with is pretty elegant.

Are both Colsons cantilevered?

Since you already have to heavily modify the Nanos, I think you could get a degree more compact with a custom two stage gearbox. Saves you some weight and packages a bit better. I guess the space savings doesn’t help much since the CIMs are sticking out past the gearboxes anyway, but hey, more room for encoders.

First thought: This is going to be impossible to make a module drive straight. A very small speed differential between the two wheels will make this turn for any reasonable driving speed gearing.

Second thought: Making a module point in the desired direction requires running the differential for a short (possibly VERY short, depending on the gearing) period of time and then stopping. This seems like it would be highly inaccurate.

Third thought: Trying to make 3 or 4 modules all point in roughly the same direction this way without mechanically connecting them seems impossible, especially given the update rates of our control system. To make it controllable for steering, you would need to gear way too low to be usable as a drivetrain.

This was a good thought exercise.

Increasing the spacing between the wheels should improve steering stability. Perhaps if they were outboard of the gearboxes things would improve enough to become usable. With a bit more adjustment, this could also be used to lower the CoG of the module even farther, though with an increase in MoI around the serve axis and a need for more real estate for the module.

If I understand correctly, that would be crab drive, not swerve. Crab allows you to translate anywhere, but not rotate the robot effectively.

Cool concept. Can the problems with steering-too-fast be somewhat alleviated by putting the wheels farther apart? Then the module gets super wide, but what if you take two of these super wide modules and build a “skateboard swerve”?

Let the thought exercise continue!:smiley:

I think this depends on how low-friction the bearing setup for the steering is. Going with a more high-friction solution (i.e. a big hunk of close fit Delrin) could provide enough resistance to prevent slight differences in speed from spinning the module. Combine that with some closed loop velocity control between the two wheels, and I think it’s a solvable problem. Not saying it isn’t more trouble than it’s worth, though.

There would be not a differential, just two cantilevered wheels. All the comments so far are spot on. A higher friction pivot bearing would help. You would probably want 3 encoders, but maybe just one would work. The control loops are real interesting. I was thinking you could have one module with minicims to get around the cim motor limit.

One idea no one talked about is putting something like a caster angle into the modules. You could shift the wheels back slightly from the centerline of the pivot. The advantage being that they might track straight easier. The disadvantage being reversing direction would need a 180 degree twist.

This was mostly a modeling exercise, but with some development, it could be a viable drivetrain.

Step files coming soon.

Do your thoughts on overall plausibility change if a non-standard wheel configuration (e.g. 3 pods in a triangle) is used?

Interesting design, one thought though:

Assuming FIRST keeps the same rules that they have the past few years, in all likelihood there will be a limit of 4-6 CIMs on a robot which would make this design as shown illegal assuming 4 wheel swerve. That said, you could just swap them out for MiniCIMs.