Ran it again, this time set the extension value to 1000 (as per my setup)
It ran with tight belts this time. Failed with low fitness value after last rectangular iteration. Was able to take tension, move to centre with Park function and raise and set Z, did an air file cut. Log attached.
Maslow-serial - 2026-05-05T083919.985.log (37.6 KB)
1 Like
Ran the v20.1 version after running LM version, with the modification #943.
Failed to get a good value on the last iteration. Log file attached.
Maslow-serial - 2026-05-06T095759.389.log (29.0 KB)
Also screen dump showing expanded serial file mid run.
Yes, David Right not Left, had my shoes on the wrong feet.
Using the normal V1.20 Find Anchors.
On this serial log, ran it again (3rd time) building from V1.20 maslow.yaml file. Didn’t give a fitness report, went to Belts Extended. Maslow was in Right down corner. I applied tension and moved to the centre, I then loaded my original version back in and ran Find Anchors. It completed with a fitness of 1.677
Find Anchor Values:
Fitness: 1.6772897
Maslow_tlX: -161.7
Maslow_tlY: 3992.4
Maslow_trX: 4552.9
Maslow_trY: 3949.9
Maslow_blX: 0.0
Maslow_blY: 0.0
Maslow_brX: 4580.3
Maslow_brY: 0.0
It then started running the Find Anchors again, probably because I had multiple windows open. I killed it
I think the closer the starting point values for the Anchor points are too actual, the more likely the fitness will be acceptable.
1 Like
I agree, it is very starting value dependent.
That is one of the things which is supposed to be a big improvement in the LM version, it should handle incorrect starting values much better…TBD if that is true though
Version 1.21 is out!
Lots of great fixes and improvements in there.
I wanted to get it out before messing with the new anchor locating algorithm in case I break things
2 Likes
@bar Suggest you put new LM Find anchors for testing purposes here (firmware and index.html.gz)
1 Like
That is a great idea. I have a couple of more changes to it that I want to make, but as soon as it’s ready for testing I will put it up.
Here is the new version for testing:
firmware.bin (1.9 MB)
index.html.gz (140.3 KB)
It should be functionally very similar to the old version from the outside, but it has an entirely new algorithm for computing the anchor point locations and the new algorithm is MUCH faster. So much faster in fact that it can now be run entirely on the machine meaning that we don’t need to depend on the calculations taking place in the browser.
That should be a big deal for reliability and consistency.
Currently it does not have an analog of the “fitness” score which we used to display. That’s a bit of an issue because it means we don’t have a way to detect if something has gone wrong physically leading to inaccurate measurements.
I am working on a fix for that now.
I would love feedback if anyone is willing to give this version a try. It’s still quite experimental, but also really exciting.
1 Like
Here is an updated version which uses an analog of the fitness function to stop the process automatically if something goes wrong:
firmware.bin (1.9 MB)
index.html.gz (140.3 KB)
Tried this with standard maslow.yaml.(v1.20) Failed after waypoint 4. Lost connection
Maslow-serial - 2026-05-13T103501.468.log (8.6 KB)
Tried with “normal” maslow.yaml Failed after waypoint 5
Maslow-serial - 2026-05-13T105152.005.log (4.5 KB)
went to Belts Extended.
1 Like
It looks like both of these are failing because it can’t find consistent measurements with error
[MSG:ERR: Measurement error, measurements are not within 2.5 mm of each other, trying again]
It looks like it’s not even making it to the new math. Do you not see that issue on the normal version?
It appears to be somehow related to the maslow.yaml file. This time I went back to an older version of the yaml file (March 31st). It failed at Waypoint 14 and went to Belts Extended. Trying again.
Maslow-serial - 2026-05-13T134146.939.log (6.6 KB)
Failed After waypoint 5. This log continues session.
Maslow-serial - 2026-05-13T135348.019.log (9.4 KB)
[MSG:INFO: Measured waypoint 5]
[MSG:DBG: Waypoint 5 coordinates: X=-49.315 Y=-1.048]
[MSG:DBG: Find Anchors LM iterations=25 bestSSR=0.547]
[MSG:ERR: Find Anchors fit failed: LM did not converge (rejections=11 lambda=1.000)]
[MSG:ERR: Find Anchors recompute failed]
[MSG:INFO: Find Anchors state reset]
[MSG:INFO: Requesting state change from Finding Anchor to Belts Extended]
[MSG:DBG: Belt positions saved to NVS: TL=2027.95 TR=1983.21 BL=1798.67 BR=1930.06 state=4]
[MSG:INFO: Succeeded]
3rd time, went to centre position, and restarted Find Anchors. WP 14
It looks like this one hit the new threshold limit. I’m going to ask the AI what that means.
[MSG:ERR: Find Anchors fit failed: anchor jump=72.526mm exceeds limit 50.000mm between recomputes]
[MSG:ERR: Find Anchors recompute failed]
[MSG:INFO: Find Anchors state reset]
[MSG:INFO: Requesting state change from Finding Anchor to Belts Extended]
Later, it looks like it would be easy to map each point to the display using the position information for each Waypoint. Getting the algorithm right is the priority now though.
1 Like
I’m working on a phone so I can’t easily post files, but here is a version which should have that error removed:
I like the idea of displaying the points in the UI
I’ll give it a go now.
1 Like
@bar Fast and no errors reported. Impressive!
No other changes to maslow.yaml
Maslow-serial - 2026-05-13T160521.755.log (16.7 KB)
1 Like
Very exciting!!! Thank you for testing it.
Here is another new version which is running the new math (on the machine) but it has some of the guard rails backed off a bit. I was having the previous version fail even in situations where I felt like it should be passing.
firmware.bin (1.9 MB)
index.html.gz (140.7 KB)
Can’t test now, off to the dentist. will have a look this afternoon
1 Like