scurve trajectory planner

@Arciera,

That's what i think we don't need. If the dogleg solver is able to find the endpoint in 3d space. Why do the tranformations
as used in the abstract?

We have already a case where the helix tail or start from a G2, G3 is used as end vector.

I see you have splitted up the gaus interpolation into x, y, z.

Edit:
Did a mayor update on the infrastructure.
- Use struct for clothoid data.
- Use seperate plot function.
- Use enum to choose a solver from the list.
- Use a central residual function used by all solvers. Where the objective function can cast types.

Why do the tranformations as used in the abstract?

Can't give you a definite answer to that but as long as we can't do it as described in the paper is seems a bit premature to want to modify the described method.

I see you have splitted up the gaus interpolation into x, y, z.

Well that's just a rough hack really that would need some streamlining but as long as nothing converges there is no point in polishing.
[edit]
And it looks like I have done it wrong having set ltot to 0

Hi guys,

looks like you are making good progress. Short question regarding the single_clothoid_fit() are you only fitting the start and endpoint or are you also trying to match the curvature etc.?

@Arciera,

The only motivation in the abstract is to align the normal n with the positive z axis direction.
The equatations for the math are already there.

- Added clothoid interpolation
- Added dynamic plot size

 

@Fabian,
looks like you are making good progress.
Yes. To be honest. I did not know we could do this at all. My math level is not that good.
The progress is fenominal. We now have at least a working solution.

We can now refine, polish code and add different solvers to the codebase. Arciera is hard
working to try out new non-linear solvers, wich are kind off trickie to set up.

The result is a G3 solution between previous and next gcode segment.
Indeed, it matches curvature.

Later on we can test more graphical in opencascade. Wich has also a great codebase
to do forward & inverse transformations.

And maybe, maybe i will code a bi-clothoid in the future. As a 4-clothoid can cause overfitting,
wich is inherent to higher degree curves.

Arciera is hard working to try out new non-linear solvers

I'm afraid that is a bit of an overstatement, I have no experience with this and also not really the time to poke around in dark any longer. This requires somebody who actually knows what they are doing, at least somewhat.

What is needed is a robust numerical solver and then the next step is the max error fitting described in the paper.

Ceres template for our codebase

Ceres example