Maslow 4 - The next generation of Maslow

Technical Details
403

Owner of a pair of Maslows (M2s actually) and a thankful backer of the Maslow 4.

A couple observations/questions- hopefully I haven’t overlooked the answers somewhere- my apologies if I have. Furthermore, I make NO claims that I am smart enough to ‘gotcha’ anyone here- bar or RomanG especially!

So…

There have been multiple mentions that the 4 can be used in a horizontal (tabletop) orientation- which opens up numerous possibilities. However, it seems to me there is a recurring design assumption that it will operate in a near upright (easel) orientation. This manifests primarily in references to top/upper and bottom/lower, or dynamics where a gravity force vector is assumed.

As an example - in update 2’s description of auto calibration it is stated that

“The order in which the belts are pulled tight is important. Here you can see that at the bottom left corner of the sheet it is important that the lower left belt is taught before the lower right belt is pulled tight when taking a measurement.”

(And I’m going to be pedantic and point out “taut” vs. “taught”- roast me on my own inevitable errors!)

This assumes the top two belts are already in tension due to gravity’s effect on the sled and thus the sled’s position is already defined by them as in the Maslow’s fundamental principles. To look at it another way- there is no reason to ‘push’ the sled (which is not possible with chains or belts) into position- it will always hang at the extent of the upper chains/belts at a particular position. So in the example above, the order of operation for the lower belts is constrained by the need to avoid pulling the sled out of this static position.

When in a tabletop orientation, this gravity-provided force is not available to move the sled- all movement has to be accomplished by pulling it into position- and this is of course the power of the Maslow 4 design. But now the order of all four belts is in question, at least for calibration where the geometry is unknown. Has this been considered in regards to the auto-calibration routine?

(I have a couple of ideas- probably closer to opinions- as to methods of addressing this, but I am curious as to others’ thoughts.)

My second question is in regards to the question of the effective ratio of the belt spools and the effect on belt tension vs. motor current.

I understand that the actual movement of the belts is measured directly by encoders, but the gearmotors’ rotation is not directly measured. As the belts are wound and unwound from the spools, the effective diameter of each spool changes, as the belt stacks up. A back-of-a-napkin calculation seems to show that this effective diameter could vary by 50% over the length of the belt.

Again, this has no effect on the accuracy of the length measurement- the encoders are measuring the belt’s movement not the spools’- but it does have an effect on the ratio between motor turns and belt movement, and thus motor current and belt tension. I have no idea how significant this would be in the actual application at hand, but I do know from experience with winches that this is a factor that can have real-world effects.

Again, thanks to both the team and the community and looking forward to this next evolution.

3 Likes
404

You are exactly correct. The automatic calibration process will need to work differently in the horizontal vs vertical setup. In the vertical setup we can rely on gravity to make sure that the upper belts are always taught :stuck_out_tongue: , while in the horizontal configuration we will need to apply tension from the lower belts to move the sled. So far when doing calibration in the horizontal setup I’ve just pulled gently on the sled to simulate gravity which works fine, but it super annoying to actually do so you can count on calibration working automatically in both the horizontal and vertical setups when we ship.

That is a great point. I hadn’t thought of that. Since we know the spool diameter and we know the length of the belt it would be possible to compute a correction factor that allows us to relate the motor current directly to the belt tension, but we are not currently doing that. That’s an interesting avenue to explore if we need it.

Excellent insights!

3 Likes
405

…and of course, once you have those two possibilities figured out, someone will have a situation where they need to mount their Maslow4 at an odd angle that is neither vertical nor horizontal!

2 Likes
406

I still say that the order is going to be wrong if you are in a different place
on the workpiece, so you need to go through more than just once to get things
tight, or (probably better) apply power to all motors at the same time.

David Lang

1 Like
407

@bar - Thank you for this entire avenue of inquiry and development, and for the kind words.

1 Like
408

@jwolter - I think that any near-horizontal angle would act as horizontal, up to the point at which the sled’s coefficient of friction with the surface allows it to slide ‘downhill’. Conversely, any near-vertical position is effectively vertical for this purpose. There is a region, I would guess somewhat closer to horizontal than forty-five degrees, where the resultant downward movement would not be reliable or fast enough to be dependably used. The question is- could the horizontal method function if there are intermittent un-commanded movements?

@dlang - “get things tight” brings up some interesting thoughts as to pre-conditions.

I have to keep in mind that it is not the length of the belts that are measured, but the change in the belts’ length. So the first potential question- does the calibration process start with the belts in a homed position and then extended under system control (or the lengths manually entered I suppose), or is the system initialized with the belts already extended and attached to the corner anchors? If it is the former, then the initialization process starts with a defined set of boundaries for both the size of its area and its position within it. If it is the latter, then either some positional parameters would have to be entered, or the system would have to make some extremely tentative initial movements to establish its environment.

From a simple geometric standpoint, there are some restrictions on the workspace and the position of the anchors;

  • The maximum possible extent of the workspace is limited by the maximum length of the belts. bar has raised this point earlier in the thread.
  • The anchors must lie beyond the workspace in both dimensions, by a distance sufficient for the two adjacent anchors to provide enough leverage to position the sled to the extent of the workspace. (Bowstring deflection of a straight tensioned line occurs at with much lower force perpendicular to the line than the force applied to the line itself.)
  • Because the intent is to avoid precision in the position of the anchors, the center of the workspace cannot be assumed to lie at the intersection of diagonals drawn through the non-adjacent anchors, nor can the X and Y axis of the workspace be assumed to be aligned with a line drawn between any two adjacent anchors.

If the process can assume that any movement within the area bounded by the anchors is ‘safe’, it would appear that it should be possible to fully determine the physical environment- both the position of the anchors and the usable workspace, from a completely unknown initial state, as long as all four belts are or can be brought to a tensioned state and the initial position is sufficiently far from any boundary to allow initial movement sufficient to roughly determine position.

I’ve got a two hour commute now, so this is what’s going to occupy my brain for a while.

2 Likes
409

This is exactly how it works. We start with the belts fully retracted to a known length of zero and then extend them under system control. The goal is to have zero user input data because it’s remarkably hard to measure things precisely enough.

2 Likes
410

@dlang - “get things tight” brings up some interesting thoughts as to pre-conditions.

I have to keep in mind that it is not the length of the belts that are
measured, but the change in the belts’ length. So the first potential
question- does the calibration process start with the belts in a homed
position and then extended under system control (or the lengths manually
entered I suppose), or is the system initialized with the belts already
extended and attached to the corner anchors?

the belts start fully retracted to the sled, you pull them out to the anchor and
then it pulls them back in.

If it is the former, then the initialization process starts with a defined set
of boundaries for both the size of its area and its position within it. If it
is the latter, then either some positional parameters would have to be
entered, or the system would have to make some extremely tentative initial
movements to establish its environment.

it has to do the movements to figure out where it is (the calibration

From a simple geometric standpoint, there are some restrictions on the workspace and the position of the anchors;

  • The maximum possible extent of the workspace is limited by the maximum
    length of the belts. bar has raised this point earlier in the thread.
  • The anchors must lie beyond the workspace in both dimensions, by a distance
    sufficient for the two adjacent anchors to provide enough leverage to position
    the sled to the extent of the workspace. (Bowstring deflection of a straight
    tensioned line occurs at with much lower force perpendicular to the line than
    the force applied to the line itself.)

correct. There is also the Z axis to consider. If the belt anchors are all at
the same level (note that in Bar’s sample frame they are below the wasteboard
level, well below the sled level) then you are further limited in how close you
can get to the anchors by the steepness of the angle of the belts from the plane
of the workpiece.

  • Because the intent is to avoid precision in the position of the anchors, the
    center of the workspace cannot be assumed to lie at the intersection of
    diagonals drawn through the non-adjacent anchors, nor can the X and Y axis of
    the workspace be assumed to be aligned with a line drawn between any two
    adjacent anchors.

correct.

If the process can assume that any movement within the area bounded by the
anchors is ‘safe’

This is not quite a valid assumption. you can have points inside the anchors
that are too close to the line between the anchors (your first point above)

and you can have points inside the boundry above that are off the workpiece so
you would end up danging in the air.

now, you could have the user manually drive the perimeter of the work area once
you have a rough calibration, and then do a multi-point fine calibration once
you know the safe area to work in.

, it would appear that it should be possible to fully
determine the physical environment- both the position of the anchors and the
usable workspace, from a completely unknown initial state, as long as all four
belts are or can be brought to a tensioned state.

Except for the fact that the usable workspace from the anchors and the
usable workspace from the supporting surface may not match, that’s what we are
hoping to get to.

And here’s a further thought to eat up your time. Can this be generalized to N
anchor points?

In theory, 3 anchor points should be enough to work. The usable work area is an
odd shape, and not well matched to 4x8 sheets of plywood. But if you are doing
this on a horizontal surface, 3 anchors can be spaced out more then the height
of a room and work.

I think that the calibration is depending on the system being overconstrained
and so may not work with only 3 anchor points. but measuring out a triangle (or
even making it equalateral) is a trival task compared to an accurate rectangle.

Similarly, if you have problems getting to the top center of a near-vertical
sheet, adding a 5th anchor above the top center would help movement in that area
and let you get closer to the anchors, reducing the overall size of the machine.

David Lang

1 Like
411

you are assuming a workpiece area and rough shape. I think we need an option
(possibly early on) to give the machine information about what the workpiece
area is (I’m thinking in terms of manually driving the perimeter)

This should be an option for people doing non-standard things, but you don’t
want the calibration falling off the workpiece into thin air because you are
trying to measure a 4x8 sheet and there’s only a 2x4 sheet there.

David Lang

2 Likes
412

I’m coming into this with no experience in calibration and don’t mind having to provide input for accuracy, but I understand the goal of fully auto.

What about landmark calibration assistance? For example, having a staple shape, colour, or image on various landmark measuring points on the sled or belts/anchors. A sensor or image taken by user can provide feedback to update positional data. Each anchor could have a symbol of a specific size so an image taken can also predict depth, rotation, and relative position to sled and anchors. Depending on the depth of software, could even map the curve and expected tension of the belts. Could this improve the calibration or is it not accurate enough/too complicated to make work well?

413

the fundemantal problem with calibration that we ran into with the earlier
maslow designs is that it’s really hard to measure accurately over significant
distances.

you can get cheap digital calipers that can measure pretty accuratly up to
~150mm, but past that it starts getting expensive fast and over ~600mm it gets
REALLY expensive.

trying to get anything accurately printed over that scale is also hard. We had
people try to get grids printed and then cameras on the maslow find the points
on the grid. The camer/maslow part worked pretty well, but they couldn’t get an
accurate grid printed without spending a huge amount of money (banner printers
aren’t that accurate as it turns out)

so the problem with your suggestion is how to accurately position the various
landmarks and tell the system where they are.

David Lang

1 Like
414

What if we could measure movement of the sled? I’m thinking of these laser mouse sensors, they must be pretty accurate, and cheap! They can sell these USB mice for like 10$, the sensor in there must be what, 2$?

Edit: I’m thinking of something like this: https://www.aliexpress.com/item/4000257811847.html?spm=a2g0n.productlist.0.0.4a45404eVdqQV3&browser_id=785d18fe4d2540c088862a97afbeb432&aff_platform=msite&m_page_id=uoqhzyftxocarjir18a5d7fb75c2326a650a2a99af&gclid=&pdp_npi=4%40dis!CAD!1.35!1.22!!!0.98!!%402102191b16937857657292867d088a!10000001050727258!sea!CA!0!A&algo_pvid=5620a392-bfba-4019-8953-8cb951f34a77

1 Like
415

We looked at those, They aren’t as accurate as you think over distance (if you
think about it, you
don’t care when using the mouse, you just move it a smidge more)

they are also very sensitive to the surface under them. They are camera based
and so require a surface with enough details for it to notice, and don’t work
well with dust.

This is why the calibration routine that they came up with for the maslow4 is
such a wonder.

David Lang

2 Likes
416

Somehow i missed the kickstarter campaign. :confused:
How can i order a maslow 4?

417

you will need to wait until the kickstarter kits are shipped, then ordering will
open again.

David Lang

1 Like
418

Thanks, good to know they were looked at.

419

The really big issue with the computer mice sensors is that they don’t handle vibrations well. The vibrations from the router make the mouse think it’s moving even when it’s not :confused:

420

I wasn’t thinking about using them during the cut, but for calibration. If you could have feedback of how much you moved and in which direction from a given belt movement, I guess it would help a lot during calibration. It probably wouldn’t have to be over a long distance either. Plywood is probably a pretty good surface for these sensors, and I don’t think anyone cuts glass with a malsow…

1 Like
421

but plastic, aluminum, particleboard with the hard layer on it, etc

David Lang

422

Belts are slow learners, it’s hard to teach them new things

2 Likes