#kdiehl
sali vom bueb
ToDos for you our friend, due 01.11.2026
- get claude max access from #eschaerer ask the sugar d....
- 1 Selfhosted Livesync get obsidian running
- 08 Remote Connection Guide
- Claude on Euler & Workflow
- install code (accept invite)
git clone git@github.com:AISCO-IDEAL/code.git
Introduction
You must check the problem: 7 thinking steps!!!
- Objective: minimise mass, while covering all holes
- Approach: global_csg, see 2026-10-22 Tim for why are not implementing PatchGrids code
- Extension: we want to bore holes and have mass placed most effectively to minimise displacement at hole 0
- Problem: the problem is problematic, displacement Displacement Losssensitivity is localised to a few holes only, most have no FEA gradient:
- Second Experiment: Cantilever Beam cantilever beam, a very standardised literature problem
- ginn used to be like this:
- then, with some improvements and the addition of latent input concatenation (we add one-dimensional z-values in range(0,1) as add. input to x/y/z values to the SDF), ginn looks like this
- csg used to look like this:
- with some improvements, we brought it to this:
- note that there are still a lot of issues with csg:
- Diversity (open issues with csg)
- holes are rarely created:
- often, networks dominate, bringing the sdf down to just 17'000 / 67'000 active parameters:
- plugs are formed, we don't know why:
- fragments are often created at stupid locations, we don't know why:
- holes are rarely created:
- We are currently deeply analysing the implemented code to then come up with a 3rd experiment that probably needs the architecture to generate sharp edges in the free design space (e.g. trailing edges, Blitzableiter, cutting tool etc)
- Read through
AGENTS.md(for coding policies) (it's on the/coderepository) - Modular Approach for code
configs ├►global_csg ├►ginn ├►bench ├►drillhole_experiment └►cantilever_experiment scripts ├►trigger_global_csg ├►trigger_ginn └►trigger_benchmark src/gurken ├►approaches/ ├►experiments/ ├►losses/ ├►sampling/ └►utils/ - Next steps: see the "dotpm" plugin: