Mecanum Drive control

[AriMindell]

We're getting ready to use a octocanum drive on our robot this year for the first time ever (very exciting!) We have a lot of past knowledge about the tank drive side, and I have thought pretty carefully through shifting, but I am inexperienced with controlling a mecanum drive.
We built a prototype mecanum bot and programmed it using some of Ether's control algorithms supplied here. We were able to get this working with regular and field-centric control, but found that due to the imperfect speed-voltage relationship, the robot does not always move in exactly the intended direction.
My first thought to fix this was to speed-control the wheels. Instead of mapping the joystick values to percent voltage of the wheels, I want to map them to percent of max speed of the wheels, using PIDF to control the speed.
Is this a viable solution to the problem? is this a common problem? what are some other solutions/implementations to improve the reliability of the robot's motion?

[SamCarlberg]

Mecanum drive is already built into WPILib

You could do PIDF velocity control for each wheel if you really wanted. But the library code works quite well and it might not be worth the time a custom PIDF control setup would take to implement correctly.

[David Lame]

In the past, we've had good success with using a gyroscope to correct the tendency to drift on mecanum drives.

I would think there would be a problem trying to use encoders to correct the wheel speeds, although I've never tried it. The problem I would anticipate is wheel slip. When using encoders for other purposes, I found there was a fair amount of wheel slip with mecanum wheels. To keep the thing driving straight, you would have to have an accurate measure of wheel speed on the ground, and slip would get in the way.

Back when I first joined as a mentor in 2014, I found a good article online about how to use the gyroscope for correction, and we implemented it that year. It involves a PID loop with the gyroscope measurement versus the desired angle for the error input. That article pointed out a few "gotchas", such as the unfortunate tendency for gyroscope chips to drift, and the interesting "kick" that can happen at the end of a turn, but it gave solutions. Unfortunately, I don't have the link to the article, but some googling should find something similar.

We had a very good programming lead that year who understood the concepts, but he was a bit sloppy. Be very, very, careful to avoid sign errors in your PID constants. If you get it wrong, you could go into what was affectionately known as the "death spiral".

[slibert]

Originally Posted by AriMindell

We're getting ready to use a octocanum drive on our robot this year for the first time ever (very exciting!) We have a lot of past knowledge about the tank drive side, and I have thought pretty carefully through shifting, but I am inexperienced with controlling a mecanum drive.
We built a prototype mecanum bot and programmed it using some of Ether's control algorithms supplied here. We were able to get this working with regular and field-centric control, but found that due to the imperfect speed-voltage relationship, the robot does not always move in exactly the intended direction.
My first thought to fix this was to speed-control the wheels. Instead of mapping the joystick values to percent voltage of the wheels, I want to map them to percent of max speed of the wheels, using PIDF to control the speed.
Is this a viable solution to the problem? is this a common problem? what are some other solutions/implementations to improve the reliability of the robot's motion?

We've found there are several levels of improvement that can be achieved to get mecanum running smoothly:

1) First and foremost, your mechanical design should strive to maintain equal weight distribution across each of the four wheels. A few years ago we purchased four digital scales just for that purpose.

2) We use the navX-MXP to provide some very cool "drive in a straight line in a given direction" and "automatically rotate to a specified angel" features. There's a robot-level "rotate-to-angle" PID rotate controller that drives this.

In teleop, the rotate PID controller can take over the Z axis of the joystick, while the driver continues to manipulate X/Y direction/magnitude via the joystick.

3) We also use velocity PIDs on each wheel, as you suggest. We found these need a feed-forward component to work at low speeds. We tune them so that we are sure the robot will move at our minimum required speed (e.g., 1"/second) for small rotations.

We found that once the velocity PIDs are tuned well, it becomes very easy to tune the rotate-to-angle PID.

These require encoders on each wheel, and we use the Greyhill 63R encoders as they are very reliable.

And we moved to the Talon SRX motor controllers last year to implement "drive a known distance" commands in autonomous.

If you're new to this, I'd focus first on 1) and then add 2) before tackling 3).

If 1) is ignored, 2) doesn't work very well (we call it the "drunken sailor" effect). And if you do 3), it makes 2) even better.

Our 2017 "Drive Mule" code that implements this is here: https://github.com/Kauaibots/FRC/blo...ems/Drive.java

There's also code in this project we developed to help w/the PID tuning in the commands directory.

[JohnGilb]

Our team took a fairly simple approach with regards to controlled mecanum drive.

1. Measure our angle relative to the field
2. Measure our instantaneous velocity

For (1), We used a highly accurate gyro. Today, the navX-MXP is a great choice.

For (2), we put three "follow wheels" on the robot to measure our velocity relative to the floor. These were small omniwheels sprung to press against the carpet with encoders on them.

We then broke mecanum drive into two components - rotation and translation.

The rotational PID attempted to keep robot heading constant unless the driver indicated a change via joysticks.

The translational PID compared the ideal velocity (from the driver's joysticks) to the actual robot X/Y velocity.

Between these two, we had a "rotational intent" and "translational intent" that were fed into the WPILib mecanum library.

This ended up being a good approach for us, since we didn't need to worry about wheel slip - the follow wheels were very accurate throughout the match. (This approach doesn't work as well when the playing field isn't level).

By applying PID on the total output of the system (i.e. how exactly is the robot moving) we were able to account for drift, slip, X/Y friction asymmetry, weight distribution, etc... with this one solution.

[slibert]

Originally Posted by JohnGilb

Our team took a fairly simple approach with regards to controlled mecanum drive.

1. Measure our angle relative to the field
2. Measure our instantaneous velocity

For (1), We used a highly accurate gyro. Today, the navX-MXP is a great choice.

For (2), we put three "follow wheels" on the robot to measure our velocity relative to the floor. These were small omniwheels sprung to press against the carpet with encoders on them.

We then broke mecanum drive into two components - rotation and translation.

The rotational PID attempted to keep robot heading constant unless the driver indicated a change via joysticks.

The translational PID compared the ideal velocity (from the driver's joysticks) to the actual robot X/Y velocity.

Between these two, we had a "rotational intent" and "translational intent" that were fed into the WPILib mecanum library.

This ended up being a good approach for us, since we didn't need to worry about wheel slip - the follow wheels were very accurate throughout the match. (This approach doesn't work as well when the playing field isn't level).

By applying PID on the total output of the system (i.e. how exactly is the robot moving) we were able to account for drift, slip, X/Y friction asymmetry, weight distribution, etc... with this one solution.

This is a really sweet solution.

Is there any mechanical design info or pictures you can post of the spring-loaded omniwheels you're using?

[JohnGilb]

Sadly, right now the only image I can find is pretty bad (attached). You can barely see 2 of the 3 follow wheels near the center of the chassis. All the spring attachment points and encoder mounts are occluded.

I'll have to poke some of the people who have CAD access to see if they can share anything. =]

[TylerHarmon]

Do not use encoders on mecanums. It will not work. Mecanum wheels slip, which completely destroys the value/quality of the ticks that your encoder returns. How many times a mecanum wheel turns cannot directly be translated into distance. Again, don't try to use encoders it will not work.

If you put a gyroscope on your robot, and you know the direction that you want to be going, you could use a PID loop to keep the robot on the heading. All you would need to do is set the PID's setpoint to the angle that the robot should be at, and that should compensate for a lot of drift.

Good luck,

-Tyler

[AriMindell]

Originally Posted by TylerHarmon

Mecanum wheels slip, which completely destroys the value/quality of the ticks that your encoder returns. How many times a mecanum wheel turns cannot directly be translated into distance. Again, don't try to use encoders it will not work.

So I agree with this assertion for trying to determine the position of the robot using encoders, but what I meant was using encoders to ensure that all of the motors are actually spinning the same speed (rather than just recieving the same voltage) for this, I think encoders can be adequate, since slipping does not come into play between the motor and the wheel, only later.

In other words, there is very little error in the encoder reading of the wheel angle relative to the robot (maybe some backlash in the gearbox) however I agree that there is a large amount of error in the encoder reading of the robot position relative to the field. Correct me if I'm wrong, but I think that since we are looking to use the first one, encoders are acceptable.

[Ether]

Originally Posted by TylerHarmon

Mecanum wheels slip, which completely destroys the value/quality of the ticks that your encoder returns. How many times a mecanum wheel turns cannot directly be translated into distance.

Whenever I see someone post "Mecanum wheels slip" I always wonder what they think that means.

[Oblarg]

Originally Posted by Ether

Whenever I see someone post "Mecanum wheels slip" I always wonder what they think that means.

Well, their effective COF is decreased by a factor of 1/sqrt(2), so I suppose they are more prone to actually slipping.

[Ether]

Originally Posted by Oblarg

Well, their effective COF is decreased by a factor of 1/sqrt(2), so I suppose they are more prone to actually slipping.

Is that what you meant, Tyler? They're more prone to slipping in a pushing match or hard acceleration? But most of the time, if you're careful, they won't slip?

Or did you mean they slip all the time, which completely destroys the value/quality of the ticks that your encoder returns, and therefore how many times a mecanum wheel turns cannot directly be translated into distance?

[TylerHarmon]

Originally Posted by AriMindell

So I agree with this assertion for trying to determine the position of the robot using encoders, but what I meant was using encoders to ensure that all of the motors are actually spinning the same speed (rather than just recieving the same voltage) for this, I think encoders can be adequate, since slipping does not come into play between the motor and the wheel, only later.

In other words, there is very little error in the encoder reading of the wheel angle relative to the robot (maybe some backlash in the gearbox) however I agree that there is a large amount of error in the encoder reading of the robot position relative to the field. Correct me if I'm wrong, but I think that since we are looking to use the first one, encoders are acceptable.

Ah, I see what you mean. I think that could work; if you are only looking at the rotations per wheel and not necessarily how far the wheel drives on the floor, you could get more useful data.

However, if your goal is to drive straight, I would suggest using a gyroscope, because even if you drive all 4 (or 8?) mecanums at the same RPM, they still will likely not operate identically to each other because of various conditions like the consistency of the floor that could cause the rollers to spin more/less.

All that said, the main use of encoders is to see how far a spinning motor or axel has moved (and at what speed), so I still can't recommend using encoders with mecanums because they simply, by design, do not give very useful data. Even if you just looked at how far they wheels turned, that data, while perhaps being accurate, would not be very useful.

[TylerHarmon]

Originally Posted by Ether

Is that what you meant, Tyler? They're more prone to slipping in a pushing match or hard acceleration? But most of the time, if you're careful, they won't slip?

Or did you mean they slip all the time, which completely destroys the value/quality of the ticks that your encoder returns, and therefore how many times a mecanum wheel turns cannot directly be translated into distance?

Hi Ether,

When I say that mecanums slip, what I mean is that the rollers on mecanums do not consistently spin. While you can control how far a mecanum wheel spins, the free-spinning rollers that make up the wheel do not regularly spin.

In both pushing matches and hard acceleration, the rollers on mecanum wheels tend to spin in irregular ways, which essentially results in an inconsistent measure of speed and position.

I think that mecanums can be very useful for strafing and maneuverability, but trying to precisely control their motion can be very difficult if not impossible because of the free-spinning rollers.

[Ether]

Hi Tyler,

Thank you for taking the time to respond to my question.

Some of the things you wrote are still a bit ambiguous to me. I think it would be enlightening to discuss it further. Would you be interested in doing that? I don't want to pester you if you don't want to do that.