Why is swerve so slow?

[Richard Wallace]

Originally Posted by AdamHeard

254 isn't even involved in EWCP either. Did they ever do a swerve, I don't recall that.

I recall a well respected 254 mentor going on record with a clear statement that sideways motion is a waste of time.

[Tom Bottiglieri]

Originally Posted by Richard Wallace

I recall a well respected 254 mentor going on record with a clear statement that sideways motion is a waste of time.

Ha! I should have appended "for our team" on the end of that comment.

We try to build robots that can complete the game challenges in the simplest, most efficient manner (in both design and implementation). We have not seen a game where our primary scoring objectives cannot be efficiently accomplished by a 6/8 wheel drive. There may be games in the future where this is no longer true, but (I'm practicing my Pat Fairbank-isms here) we'll burn that bridge when we get to it. We've talked about building one as an academic exercise but I think there is enough community experience on the subject that we could put something together, if we needed to, without a homegrown proof of concept.

Now for other teams, this could be different. Swerve drives are incredibly cool and building them definitely inspires a lot of students every year. It's a hard engineering challenge which some people may want to try out, and that is awesome. We just concentrate on other things.

[asid61]

Originally Posted by Michael Hill

Just throwing this out there: I've got a drivetrain calculator spreadsheet on here.

http://www.chiefdelphi.com/media/papers/3038

It used Ether's algorithms, but it's in Excel format. There are some pretty graphs as well.

Even this shows that a 11fps drive will go only ~6" farther than a 21fps drive maximum. after about 3/4 of a second, the 21fps drive will go farther. This is with 4 cims.

Originally Posted by Tom Bottiglieri

Ha! I should have appended "for our team" on the end of that comment.

We try to build robots that can complete the game challenges in the simplest, most efficient manner (in both design and implementation). We have not seen a game where our primary scoring objectives cannot be efficiently accomplished by a 6/8 wheel drive. There may be games in the future where this is no longer true, but (I'm practicing my Pat Fairbank-isms here) we'll burn that bridge when we get to it. We've talked about building one as an academic exercise but I think there is enough community experience on the subject that we could put something together, if we needed to, without a homegrown proof of concept.

Now for other teams, this could be different. Swerve drives are incredibly cool and building them definitely inspires a lot of students every year. It's a hard engineering challenge which some people may want to try out, and that is awesome. We just concentrate on other things.

So 254 has never actually built a swerve drive then?

[ekapalka]

If we ever built a field centric swerve, I would almost certainly allow a drive mode that allows quick momentum saving direction changes. (Maybe a single button click? Maybe a press and hold button that locks 'forward' to the direction of your last translate command then allows throttle/turn steering?)

Originally Posted by AdamHeard

16 does this and it's why everyone thinks they are so fast.

Could you or someone else elaborate? I'm not entirely sure I understand how this works

[AdamHeard]

Originally Posted by ekapalka

Could you or someone else elaborate? I'm not entirely sure I understand how this works

Essentially their primary drive mode is Monster Truck steering. It's a traditional omnidirectional swerve derivation (like the type Ether posted) but with strafe locked out.

It will only ever let you drive smooth splines essentially, so you're always carrying your speed with you. You can still do very tight radius (and at no speed zero radius) turns.

They then have multiple (maybe just one?) other modes they toggle into for strafing variants.

16 could explain better, but I think I got the jist of it. I feel video of their 2012 bot shows this pretty well.

[Jefferson]

Originally Posted by ekapalka

Could you or someone else elaborate? I'm not entirely sure I understand how this works

We are in 4-wheel (Ackermann?) steering that is NOT field-centric by default. In that mode we maintain a lot of our momentum because the turns are very smooth. When the driver needs to make a quick maneuver, he pulls (and holds) the trigger on the joystick to enable crab that IS field-centric from the orientation of the robot when the trigger was pulled. The driver has the x and y motion on one joystick and a twist motion on the other. When the trigger is released, the driver is back in 4-wheel steering mode.

[Jefferson]

Originally Posted by AdamHeard

Essentially their primary drive mode is Monster Truck steering. It's a traditional omnidirectional swerve derivation (like the type Ether posted) but with strafe locked out.

It will only ever let you drive smooth splines essentially, so you're always carrying your speed with you. You can still do very tight radius (and at no speed zero radius) turns.

They then have multiple (maybe just one?) other modes they toggle into for strafing variants.

16 could explain better, but I think I got the jist of it. I feel video of their 2012 bot shows this pretty well.

Well said.

We have toyed with other driving/aiming modes, but they are just variants of those two basic functions.

[cjl2625]

Originally Posted by AdamHeard

Essentially their primary drive mode is Monster Truck steering. It's a traditional omnidirectional swerve derivation (like the type Ether posted) but with strafe locked out. .

What do you mean by monster truck steering?

[asid61]

Originally Posted by Jefferson

We are in 4-wheel (Ackermann?) steering that is NOT field-centric by default. In that mode we maintain a lot of our momentum because the turns are very smooth. When the driver needs to make a quick maneuver, he pulls (and holds) the trigger on the joystick to enable crab that IS field-centric from the orientation of the robot when the trigger was pulled. The driver has the x and y motion on one joystick and a twist motion on the other. When the trigger is released, the driver is back in 4-wheel steering mode.

That is so weird. Man, that must take a lot of practice to get used to!

[AdamHeard]

List of reasons swerve is "slower" than a traditional skid steer. Reasons may not be valid for all swerve variants.

-Likely has more stages of power transmission, and if using bevel gears has more inefficient power transmission. A good skid-steer is typically 2-3 stages of decent spur gears. A swerve could be 3-4 stages easy. Motor in module variants are an exception. This changes a drive efficiency from a possibly 90% to maybe around 70% (made up numbers!)

-For the same reasons as above, there is likely more rotating mass (inertia) in the drive to accelerate, this could easily be a few percentage points slowdown.

-Steering error. The odds of all your modules pointing perfect is low, there are likely appreciable losses due to friction here.

-Steering time. It takes time to steer the modules to where they want to go, and teams have different methods for dealing with this. Some assume it's instantaneous as they gear it fast, but this leads to a lot of the above error on the way there, which wastes battery and heats the motors more. Some teams turn power down until wheels are pointing where they want to go, and this also obviously adds a delay as the drive must wait before it starts really going.

-Weight Transfer. Depending on the exact gearing, CG, overall robot geometry, tread, etc... it's possible that during initial acceleration enough weight is transferred such that half your wheels are now breaking traction. Even if this happened for .1-.2 s it would appreciably slow you down. This is only valid for independently powered modules.

-Driving style/Control Scheme. Except for 16, most famous swerves have been driven by point to point drivers. As Tom pointed out top drivers on skid steer are forced to learn to carry their momentum to be quick. Many swerves don't do this. If we swerve again, we will spend time looking at this.

There are more reasons, and if I think of them I'll edit and add them.

None of the above are huge, but a few small things could easily add up to make something seem/feel a bit slower. Maybe if I get a good block of time I'll model the above and see if I can point out a difference.

[Jefferson]

Originally Posted by asid61

That is so weird. Man, that must take a lot of practice to get used to!

JTN could comment better about the difficulty of learning it (I'm a pretty terrible driver), but it's probably not as difficult as you think. You just have to put yourself "inside" the robot. The tough part comes when you change orientation (twist) in crab. You have to keep up with where the "front" of the robot is/was.

Like most robots, it's not hard to learn the basics, but it takes hours and hours of practice to get good. There's no time to stop and think behind he glass.

[Jefferson]

Originally Posted by cjl2625

What do you mean by monster truck steering?

You have gas (speed) and steering angle inputs. All four wheels turn to create the desired robot turn. Wheel stop turns will spin about the center of the robot with no translation.

[cjl2625]

Originally Posted by Jefferson

You have gas (speed) and steering angle inputs. All four wheels turn to create the desired robot turn. Wheel stop turns will spin about the center of the robot with no translation.

Oh ok, so the rotation point is at the center of the robot.
I didn't know monster trucks drove like that

[Jefferson]

Originally Posted by AdamHeard

-Steering time. It takes time to steer the modules to where they want to go, and teams have different methods for dealing with this. Some assume it's instantaneous as they gear it fast, but this leads to a lot of the above error on the way there, which wastes battery and heats the motors more. Some teams turn power down until wheels are pointing where they want to go, and this also obviously adds a delay as the drive must wait before it starts really going.

I'll say we assume they are correct and taking off. The modules are actually really fast and we always take the shortest route to the desired angle. This requires inverting the drive motor direction if necessary.
We noticed our drivetrain chewed up batteries much faster when we swapped the Denso window motors for the AM 9012.

[Meredith Novak]

Originally Posted by Jefferson

JTN could comment better about the difficulty of learning it (I'm a pretty terrible driver)

You killed that chair...