Anushrut Jignasu

Graduate student at Iowa State; Computer Vision, Graphics, Deep Learning

[Anushrut Jignasu @ajignasu]

@eliec_ @Dylan_Morri @afshawnl has done some good work on this front

[Anushrut Jignasu @ajignasu]

Huge thanks to Daniele Grandi for his mentorship and collaboration. This work grew out of my internship at @autodesk.

Paper: https://t.co/0lmj3SviPL
arXiv: https://t.co/i8AYwc4jwI
Code + dataset: https://t.co/S64nlbANLt

[Anushrut Jignasu @ajignasu]

Our paper, "Linkify: Learning from Interface-Augmented Assembly Graphs" is now out in the CAD special issue on AI for Geometric and Physical Modeling in Design and Manufacturing!

We recomputed high-fidelity interface geometry for the Fusion 360 Gallery Assembly dataset, correcting missing and erroneous contacts. We then show that this interface geometry (without relying on joint types) provides meaningful signal for understanding how parts function together. Using these interfaces we study context-aware part-retrieval in mechanical assemblies using GNNs.

[em m0shouris @emm0sh]

here's my conversation with Carl Bass. we talked about CAD, engineering software, AI, Autodesk and organizations

i hope you enjoy it as much as i did

[Anushrut Jignasu @ajignasu]

@connorkapoor @afshawnl Yeah we have :)

[Anushrut Jignasu @ajignasu]

@mattfreed there are a couple of solid approaches out there to tokenize parts (autobrep's unified geometry+topology, UV-grid etc). but not enough focus on preserving manufacturing intent and assembly dynamics

[Anushrut Jignasu @ajignasu]

Connor’s solving some interesting geometry/physics problems with metal casting! Highly recommend reaching out

[Connor Kapoor @connorkapoor]

im hiring again. MechE/Builders

who you are:

> obsessed with how things are made
> going to build something before the CAD is complete
> interested in reindustrializing America

what you get:

> highly competitive salary
> equity
> radical ownership over a problem, and the autonomy to solve it

Shoot me a dm with a link to your favorite project you've built, or reach out over email with the same.

[Anushrut Jignasu @ajignasu]

@amypretzel majority of the generative CAD models are learning geometry distributions and not physics. but i think working in program space (sequences of ops) can address this

[Anushrut Jignasu @ajignasu]

@afshawnl Hybrid reps might be used. Brep Distance Functions (https://t.co/od0Bpl8KNm) have a lot more room for improvement

[Anushrut Jignasu @ajignasu]

@aphysicist Yes! Happy to help with it as well

[Anushrut Jignasu @ajignasu]

@vincesitzmann nice blog post! we're already on the path to this with slice-100k (https://t.co/xJJcLuuC40). 100k+ paired G-code, meshes & renderings for finetuning on the printer interface.

[Anushrut Jignasu @ajignasu]

@devonkcopeland @amypretzel @d4ferris Our viewer is limited to ~80-90 parts

[Anushrut Jignasu @ajignasu]

@gkopanas They could be seen as a DSL, but its really just an API wrapper. Not built for gradient descent

[Anushrut Jignasu @ajignasu]

@yacineMTB We’ve had good results with 3-axis aerosol jet printing using a conductive ink: https://t.co/FGCpdfzusH

Still lots of room to push this forward on in-the-wild geometries

[Anushrut Jignasu @ajignasu]

@zachdive @adamdotnew normal map maxxing

[Anushrut Jignasu @ajignasu]

Assembly simulation and auto-fix loops still seem like open territory for text-to-CAD

[Anushrut Jignasu @ajignasu]

@AlexanderLong @ChaminHewa Let’s go @ChaminHewa!

[Anushrut Jignasu @ajignasu]

@yongyuanxi The adapter layers trick to project CAD tokens to the LLMs embedding space is kinda similar to how LLaVA uses projection layers to align visual features with text embeddings