Disadvantages of swerve/crab

[davidthefat]

Isn't it generally accepted that a more complex system fails even harder? More moving parts allow for more chances of something going wrong? I believe that is a serious flaw in performance.

[evoluti1]

Isn't it generally accepted that a more complex system fails even harder?

NASA even has a phrase for this: "The higher the tech, the bigger the wreck." It might be a little late in the game to be thinking about a crab drive at this point. My team wanted to do crab drive this summer, but earlier this month we realized we simply didn't have time to finish and refocused our efforts towards a WCD.

Your team may be different, of course. If you have the financial/technical/mentor resources to make a crab drive happen during season, go for it. But from what I've seen, only very elite teams have those kinds of resources, and even most of them prefer to stick with something simpler.

You have to remember, this is real life. Of course, a "perfect" crab drive beats any other drive on the field. But the level of quality/perfection that makes a crab drive competitive is very difficult to achieve in the real world.

[Jared Russell]

* Often less intuitive to drive than skid steer robots
* There is latency to all changes of direction because module steering is never instantaneous
* Geometry of swerve (assuming normal bumper zone rules) means that wheels cannot be put as far out to the edges of the chassis as in a skid steer case (may make being tipped easier)
* On uneven terrain, one or more motors may be unable to deliver power to the ground
* Unless you build an all-wheels-driven-and-steered-independently ("unicorn") swerve, there will always be certain combinations of manuevers that you cannot make.

[Ether]

Originally Posted by Akash Rastogi

Even with a full practice swerve to play with and program, we had only just started to figure out the best controls for the swerve WHILE in Atlanta that year.

In the above context, "the best controls" sounds like you are referring to the driver interface. What driver interface did you finally decide was the best for your team?

It is a HUGE learning curve for even the best programmers. (we had students and programming mentors working on it)... even with the mechanics pretty much figured out for us through COTS parts, the programming and sensory involved took a long long time to figure out.

Could you elaborate just a bit and list the top three (or whatever) things that made the learning curve huge and which took a long time to figure out? It might be helpful for other teams on the verge of making a similar decision.

[sgreco]

Originally Posted by Akash Rastogi

Even with a full practice swerve to play with and program, we had only just started to figure out the best controls for the swerve WHILE in Atlanta that year.

That's actually pretty quick as far as I'm concerned. 2079 built a prototype swerve before the 2008 season, we build a swerve drive for the 2008 season. We made our third iteration for the 2009 season. We just started to get it where we wanted it at the end our second regional in 2009. They have since ditched the idea of swerve do to complexity and lack of a fully intuitive algorithm for controls.

The real thing is that swerve is just harder to make well. It's not better or worse than any other drive train.

[John]

Theoretically, if you had unlimited resources, unlimited manufacturing time, unlimited programming time, and unlimited driver practice time, to build a "perfect" swerve drive, it would be clearly better than any other drive train. I would consider a "perfect swerve drive to be: all wheels independently powered, steered, shifting, never breaks , low weight, very low delays in turning the wheels, driver can intuitively control all functions, no repercussions on other systems.

Its pushing strength would be as good as a low speed skid steer,
its speed would be as good as a high speed skid steer,
its acceleration would be as good as any robot of similar weight,
its omnidirectional ability would be as good as mecanum/omniwheel drive

However, any real team does not have unlimited resources and time. As a result, most swerve drives do not achieve that performance. Some teams have gotten close, and these teams have generally been very successful, (148 in 2008, 111 in 2009). The problems with a swerve have to do with the real challenges of making one work like it is supposed to without massive tradeoffs, and not with any inherent disadvantage to swerve.

[Ether]

Originally Posted by John

Theoretically, if you had unlimited resources, unlimited manufacturing time, unlimited programming time, and unlimited driver practice time, to build a "perfect" swerve drive, it would be clearly better than any other drive train.

You didn't mention anything about waiving FRC rules, so I assume you are still limited to the legal type and number of motors.

its omnidirectional ability would be as good as mecanum/omniwheel drive

I think the jury is still out on whether it is possible to steer the wheels on a swerve fast enough to make swerve "omnidirectional ability" as good as a well-built omni or mec.

[Siri]

What Jared said.

Also, FIRST has literal strategies that jack-of-all-trades swerve just isn't the best for. Of course, if you want to master a single "trade", swerve isn't necessary. But even if you want more, swerve gives you two things--reasonable holonomic-ness and reasonable traction--all the time. If you want, say, better holonomic drive sometimes and unmovable traction other times, something like switchable wheels/Octocanum could be better. Etcetera.

Originally Posted by Akash Rastogi

What is impractical about making a multi-speed swerve?

Depends. If you're doing a limited swerve/non-coaxial, adding a shifter isn't so hard. Maybe heavy, but at least you're not rotating it and probably don't need so many. In a co-axial, you're driving each wheel independently and rotating the drive motor/gearbox with the module. Adding a shifting mechanism gets really big and heavy (and expensive and complicated) really quickly. We do all our gear ratio adjustment via belt pulleys and planetary gearboxes. After two years of working on the drive, "impractical" still seems like a pretty good word for trying to mount an AM Supershifter on this, but maybe there are more streamlined shifters?

That said, we tend to win our pushing contests.

Originally Posted by AlexH

Only effective if you have a good driver who can capitalize on the directional freedom that a swerve offers.

Moreover, only if you have enough good drive time with your chassis to get this capitalization experience. We've done poorly in this area for the first two years, but now that we have essentially optimized hardware, getting-close programming and excited drivers, this is huge.

Originally Posted by Aren_Hill

I've got a few concepts that could hit < 35lbs for complete chassis with completely independent module power and steering, so weights only really an issue if you aren't trying hard enough

Wow, that's great! Our modules are at 9.1lb each with a 14.1lb chassis weldment. How'd you get it to 35lb total?

[gblake]

One obvious disadvantage is the endless debate they engender.

One clear advantage is the endless debate they engender.



Blake

[Akash Rastogi]

Originally Posted by Ether

In the above context, "the best controls" sounds like you are referring to the driver interface. What driver interface did you finally decide was the best for your team?

Sure thing! Best is the wrong word too, "better" is the correct choice. We started off with a single 3D Pro joystick to control the drive system, but did not get a chance to figure something else out until later in the season which was using a joystick for throttle and then used the 3d joystick for rotation. It isn't best, but it was as far as we got before we decided to put off swerve until future offseason projects. But even then, we ended up using the override button on the joysticks which turned it into a 4 wheel tank drive most of the time.

Originally Posted by Ether

Could you elaborate just a bit and list the top three (or whatever) things that made the learning curve huge and which took a long time to figure out? It might be helpful for other teams on the verge of making a similar decision.

I'd say some of the biggest issues we had were 1. being dumb and not using a 1:1 reduction for the rotation, 2. getting the modules perfectly synched, and 3. realizing too late that we should have used higher quality sensors than we did.

Big problem was that our drivers expected a perfect response time from the modules everytime they wanted to move somewhere, but our gearing and sensors significantly delayed response time.

Originally Posted by Akash Rastogi

Big problem was that our drivers expected a perfect response time from the modules everytime they wanted to move somewhere, but our gearing and sensors significantly delayed response time.

How significant? Was it like major lag between controller input and movement, or more like a small yet noticeable change between input and movement?

Also, what gearing did you use, and which sensors? Which would you recommend of both?

[Akash Rastogi]

Originally Posted by SuperNerd256

How significant? Was it like major lag between controller input and movement, or more like a small yet noticeable change between input and movement?

Also, what gearing did you use, and which sensors? Which would you recommend of both?

It did what seemed liked a small change between input and movement sometimes, but later on that was fixed. The delay later on was, to me, significant because I did not know what to expect. It was just not instant as we all wanted, and I think this also happens to a lot of people when they first try swerve. The team might be under the impression that they'll get prefect turning and response from the get-go. However, with optimized systems such as 973, 111, 16, I suspect they see the kinds of results we expected: instant response (or close to).

As for the gearing, I'd have to go back to the CAD as I don't recall. The CAD should be somewhere on the Autodesk website from 2010 if they still archive that stuff.

[Aren_Hill]

Originally Posted by Siri

Wow, that's great! Our modules are at 9.1lb each with a 14.1lb chassis weldment. How'd you get it to 35lb total?

Well over thanksgiving I rebuilt 1625's 2010 practice chassis (was 6wd swerve) into an 8 motor independent swerve, that came in a 41lbs,

Mainly its just sitting there staring at a concept going "how do i get rid of that....and why is that even there?" when looking at parts

current one i have almost ready to go is 6lbs flat per module (including CIM and steering motor)

[Siri]

Originally Posted by Aren_Hill

Mainly its just sitting there staring at a concept going "how do i get rid of that....and why is that even there?" when looking at parts

current one i have almost ready to go is 6lbs flat per module (including CIM and steering motor)

That's awesome! Great job. Do you have CAD or photos?

[Aren_Hill]

Originally Posted by Siri

That's awesome! Great job. Do you have CAD or photos?

That ones sitting nice and quiet in the back pocket till it gets used