@dlang Fitness of 0.95 with the 1500x1000 calibration size and interstitial firmware 0.87.1.
Maslow-serial (9).log (21.2 KB)
I ran it again with lower tension, but I think you need to reset the machine between calibrations as it didn’t output any results which was odd. I’ll run it again today most likely then try to jog and measure.
FWIW, the flex measurement was cut in about half with 1000 vs 1300 retraction force (from 1 & 4mm to 0.5 & 2mm).
Yeah, this is a known bug. There is something funky in the state machine which isn’t getting reset
I’ve noticed some other state issues as well, like uploading a new yaml file resulted in a stale state until I restarted the machine. I’ve also seen stale coordinates when jogging which was really odd.
After the shorter calibration size and lowering the calibration tension, the X axis is actually worse by a few mil. I checked the left side and it’s within 1-2 mill of the right in terms of how far off each point is.
Might increase the calibration size to what I’m actually trying to cut and see what happens. I could imagine you’d get a higher fitness from a smaller grid, but that might not actually help in the way of mapping out the space. I’ll report back when this test is done.
I agree, the fitness number isn’t really indicative of real world performance, just how well the system thinks it is solving for the anchor points. If we just feed the system 3 measurements it will find an infinite fitness, but not because it’s perfect, just because it doesn’t have enough data to understand that it’s not perfect.
That being said it seems like there is something going on with the outer data points always being worse than the inner data points which makes me think that there is something we aren’t taking into account correctly. I don’t have a good theory of what that is yet though
Let me know if you come up with any theories I can test and I’d be happy to try them out.
Hopefully I’ll be able to run a calibration today bringing the spindle within 2" of the edges to replicate how I’d like to cut. I had to set up some overhang guides yesterday and didn’t get to running the calibration.
Brainstorming ideas:
Are the out points off further because they are calibration measured first ?
Is there any way to determine if it is the shorter belts or longer belts that are off? For example longer belts may stretch more.
Is the math expecting the anchor point to be theoretically linear but in reality the anchor eyelet aren’t.
For debug Could we manually jog Maslow to a known dimensions, ie 4ft x 4ft and record corners positions and then use as additional points to calibrate against.
Dano wrote:
Are the out points off further because they are calibration measured first ?
it computes the inner points first.
Is there any way to determine if it is the shorter belts or longer belts that are off? For example longer belts may stretch more.
we haven’t spotted a pattern
Is the math expecting the anchor point to be theoretically linear but in reality the anchor eyelet aren’t.
it moves from one point to the next, at each point it tightens all 4 belts and
records how long they are. As it finishes each pass (a square of points), it
runs those measuarements through an algorithm to try and figure out where the
anchors are. then with the revised anchor points, it moves on to the next layer
of points and repeats.
For debug Could we manually jog Maslow to a known dimensions, ie 4ft x 4ft and record corners positions and then use as additional points to calibrate against.
In theory we could, in practice, with the older maslow, we found that defining
‘known points’ at large scale was much harder than it seems to be.
David Lang
Able to make some progress today!
Cranked the calibration grid out as far as possible, within about 2.5" of the edges on the X axis and maybe 1-2" of the Y axis edges. Grid size I used was 2300 x 1117. On the initial attempt the arms collided with the machine and calibration failed so I had to move the corner anchors out a few inches with some block extensions. This worked despite the arms still colliding with the machine on the far left and right sides. Barely made it by the calibration threshold with 0.456 or something.
Results:
Quickly jogged 800mm to the right and improved from -5mm to -1mm. Homed and jogged 1200mm to the right and improved from -10mm to -4mm! Note: this far out the arm motors still collide with the top brace of the machine and cause the longer top belt to go slack (see pics). But huge improvement!
Take aways
Next Steps
Pictures & log files
for those interested
Motors hitting the frame and causing TL to go slack. Can also see the extended belt anchors.
Calibration log for @dlang:
Maslow-serial (10).log (22.1 KB)
YAML file for those curious on machine size and calibration settings:
maslow (4).yaml (5.3 KB)
play around with the calculator I setup at
http://lang.hm/maslow/maslow4_frame.html
it shows yellow areas to the side/top where adjacent arms hit and slightly
different yellow areas in the corners where opposite arms will hit.
It even lets you tinker with what the angles are where it will hit.
for the corners, one idea I have is to cut off the top bracket for the lead
screws and put a couple mm of spacer under the bottom arm, allowing the arms to
get closer to vertical (that’s the ‘clipped’ setting for the angles on that
page)
I also put together an idea for a version of the maslow that uses the lead
screws as the vertical guides. I don’t know if the 8mm leadscrews would be stiff
enough.
I haven’t tried building this yet
David LAng
Interesting idea to just clip the top support. I have an extra as an old one cracked but is still functional so I’ll give this a try. @dlang have you actually tried this and successfully made cuts? From your calculator this seems like it might fix the X axis issue by removing the interference on the L and R sides of the workpiece.
Also, can you elaborate on this? Not sure I follow. Guessing BL contacts the bottom Z motor somehow?
put a couple mm of spacer under the bottom arm
Kyle wrote:
Interesting idea to just clip the top support. I have an extra as an old one cracked but is still functional so I’ll give this a try. @dlang have you actually tried this and successfully made cuts? From your calculator this seems like it might fix the X axis issue by removing the interference on the L and R sides of the workpiece.
No, I have not tried it. I have one that’s broke, but not cut off yet
Also, can you elaborate on this? Not sure I follow. Guessing BL contacts the bottom Z motor somehow?
put a couple mm of spacer under the bottom arm
yes, there is a bulge under the arm that will contact the bottom clamp
David Lang
Okay cool. I’ll be the guinea pig for this one and lop the supports off. I’ll report back when I can 
Playing with the calculator with the intention of extending my anchors to get good results edge-to-edge on a 4’ x 8’ workpiece, it seems that if I could space my anchors at 4400mm (174") wide and 3200mm (126") high, I should be able to navigate corner-to-corner and edge-to-edge without any arm hitting. Am I understanding correctly?
I’ve got an idea for incorporating stiff anchor extensions to allow me to set anchors like that without compromising frame stiffness.
This is what I was thinking as well, but on a vertical frame you’d have to move each anchor over 26" past the 10’x8’ frame on the X axis alone, which is pretty massive. I’m hoping that modifying the maslow support will allow me to keep a small frame size.
If you have a horizontal setup, it’s not as big of a deal.
Aloha,
I built a 12’ x 8’3" frame…my anchors would be about 15" beyond the corners (so the extensions would be maybe 18" in each dimension.
I have an idea in my head for replacing my 3/4" plywood top and bottom corner braces with braces that extend into the frame space and out to anchor supports. I haven’t modeled it yet.
David Negaard wrote:
Playing with the calculator with the intention of extending my anchors to get good results edge-to-edge on a 4’ x 8’ workpiece, it seems that if I could space my anchors at 4400mm (174") wide and 3200mm (126") high, I should be able to navigate corner-to-corner and edge-to-edge without any arm hitting. Am I understanding correctly?
exactly
I’ve got an idea for incorporating stiff anchor extensions to allow me to set
anchors like that without compromising frame stiffness.
right now the firmware does not support extensions, but that size frame need 4m
of belt and you have about 4.4m of belt, so you should be good without it.
we don’t currently take into account belts stretching, but we know from earlier
maslow versions that if the anchors shift under tension it throws off the
calculations, thus the worry about the frame being ridgid enough.
to be sure, have your camera record a video of the corners from the side, on a
tripod (not handheld) as you work, then play back the video at high speed and
you will see any movement that takes place.
David Lang
@dlang top mount chopped and looks great! Might be able to calibrate later tonight or tomorrow.
Kyle wrote:
@dlang top mount chopped and looks great! Might be able to calibrate later tonight or tomorrow.
Two things
can you try to measure the angle to see how close to vertical the arm can go?
What I have in my tools is based on the CAD model I have of the machine, I’m
curios about how accurate that is.
the risk of this approach is that there will be more flex between the router
and the sled. I don’t think there will be, but it’s something to watch out for.
David Lang
The angle of the belt? I have a digital angle finder and could use that to see how it relates to the ground. Not sure how level the frame is though, might have to measure that and factor it in too.
