wilcrofter / crust Goto Github PK

Taking the cue from another project, and not getting to things as quickly as I'd like, I'll post tentative plans here with "incidental notes" tags. That way, I can avoid pestering everyone with emails and have github pester everyone instead.

Links are a data structure meant to specify relative positions of two bodies. This is OK since links can represent putative alignments as read from robotic arms and calibration. However alignments with respect to an absolute reference frame will be necessary to represent physical reality.

The 2D Fourier transform of a function of two variables, b(z,x), can be theoretically reconstructed from 1D Fourier transforms of certain line integrals. If, for example, you have the integral of b(z,x)dz, for every x, you obtain a function, f(x). The 1D Fourier transform of f(x) is the "slice" of the 2D Fourier transform which you get by setting the z spatial frequency to zero.

I'm not sure, but I imagine some variation of this technique might apply. I know it is fast but I think it is sensitive to noise (despite that the Fourier transform preserves power,) so I want to understand noise sources before getting into it.

Not having a UR5 at home (regrettably,) we are prototyping a simulator as the closest practical substitute. Since the real thing involves rigid bodies and rigid body motion, an object model would seem to be the right approach. The R language offers 3 formal object-oriented systems, S3, S4, and Reference Classes, plus various informal possibilities such as using closures.

Initially, we've decided to use an informal system. In R, functions and scopes are first class objects which can be manipulated like data. A function maintains a reference to the lexical scope, or environment in which it was defined. That environment can contain both data and other functions, hence provides an easy, if not particularly elegant, representation of an object with fields (data) and methods (functions.)

Programming interfaces evolve with use, so the system will probably change. When it's reasonably stable, we'll bundle it for easy installation as an R package on github using drat.