Enginuity Labs

Blog

Euclid-1: The World's First Purpose-Built Physical Design Model

Euclid-1 is the world’s first purpose-built physical design model. It is trained with RLFPC (Reinforcement Learning for Function and Physics Calibration), which rewards a part for working: doing its job, holding up in simulation and being manufacturable, not for looking right.

01

Reward Modelling

There is currently no real metric that tell a model what “good” looks like. Almost every CAD benchmark today answers that with geometry - IoU or Chamfer distance against a single reference shape.

This reward measures the wrong thing. It rewards a model for occupying the reference envelope, even if the feature that gives the part its job is missing. It gives no gradient toward real engineering details - a bolt hole 2 mm off barely moves IoU, yet the part no longer assembles. A sturdy chair that is slightly larger in some dimensions gets penalized, while a weak chair that overlaps the ground truth gets promoted. And because there is only one reference, every valid alternative design is marked wrong for being different.

Training a model on that signal teaches it to copy shapes and give beautiful renders - it does not learn engineering.

02

RLFPC

We built RLFPC to fix the reward. The reward is calibrated on accomplishing the function, proven through simulation and manufacturability. Running it naively puts a CAD kernel and a simulator inside every single training rollout, so we built RLFPC to optimize for compute.

Every part gets its own rubric, leveraging the EDSX representation, generated from its requirements: what the part has to do, the dimensions, interfaces and tolerances it has to hold, and how well it was constructed. That rubric is graded together with checks on the built part itself - it regenerates cleanly in the kernel, holds its loads in simulation and can be made by its stated process. The judges never see the reference part. Any construction that meets the requirements earns full reward, which means the same reward that trains Euclid-1 to rebuild known parts carries straight over to designing new ones from requirements alone.

03

Results

We evaluate on ELCDB, our CAD design benchmark: a diverse dataset of complex parts - cast engine cylinders, helical gears, pump bodies, suspension arms - each graded on function, specification and construction quality by a panel of model judges with the reference part withheld, then reviewed by engineering experts. Every model gets the same brief, the same reference images and the same tools, and builds in the same live kernel. The harness for every run was the Archimedes agent in Enginuity Design Studio.

ModelELCDB (Functionality)Cost per task (USD)
Euclid-171.1$1.60
GPT-6 Astra47.6$5.30
Fable 5.138.0$2.80
Inkling (base)7.3n/a

Euclid-1 goes from 7.3 to 71.1 over its base in its functionality scores - about 50% above GPT-6 Astra and nearly double Fable 5.1. It is also the cheapest to run, at $1.60 a task: 43% below Fable 5.1, the next cheapest, and 70% below GPT-6 Astra.

That jump is RLFPC at work. Inkling, the base, is a poor designer on its own: it scores 7.3 on ELCDB, and on the sample task below it builds a plain rectangular plate a thousand times too large. Post-trained with RLFPC, the same base becomes Euclid-1 and scores 71.1, ahead of every frontier model we tested.

04

Sample Task Comparison

One task from ELCDB: an aluminium lower suspension arm, the load-carrying link between two chassis pivots and the wheel-side joint, 231 faces in the source part. A triangulated frame with two relief bays carries the load between two coaxial pivot pads and a threaded wheel-joint boss, with an auxiliary mounting boss, three reinforcing ribs and four levels of blends. Every model received the same written brief and three reference views of the part:

Model an aluminium lower suspension arm as one body, functioning as the load-carrying suspension link between chassis pivots and the wheel/upright joint. Hold the overall envelope to 42 × 9.5 × 41 mm and the main structural thickness to 6.35 mm. The paired coaxial pivot pads are Ø8 mm × 8 mm, set 25.4 mm apart, with a continuous Ø3.175 mm through-bore defining their common pivot axis. Make all three reinforcing ribs 2.54 mm thick. The opposite joint boss is Ø9.5 mm and 9.8 mm long. Preserve the load-path blends: principal radii 12.7 and 8 mm, secondary rounds 1.6 and 0.8 mm, and a 0.4 mm, 45-degree chamfer.

The source part, top-front-right isometric in SOLIDWORKS
The source part, top-front-right isometric in SOLIDWORKS
What each model built, top-front-right isometric in SOLIDWORKS
What each model built, top-front-right isometric in SOLIDWORKS

The judges grade each part against 24 function claims and 30 specification claims written from the brief, plus a separate construction-quality score. The final score weights them 50, 20 and 30, and each layer is discounted by the share of its claims that were missed, so a part that looks close but misses many details still scores low.

Euclid-1 built the triangulated frame, both relief bays, the coaxial pads, all three ribs and the blend hierarchy, and its part is within 2 mm³ of the source volume of 3,731 mm³. Its misses are all partial: the wheel-joint and auxiliary holes are tap-drill holes with no modelled thread, one rib’s free edge is left sharp, and some rib roots use 0.8 or 0.4 mm rounds where the brief calls for 1.6.

GPT-6 Astra met the specification almost as closely as Euclid-1, but missed 14 of the 24 function claims. It left out the diagonal member, so the arm has one open window instead of a triangulated frame with two bays. The auxiliary mounting boss, its threaded hole and the rib that ties it into the arm are missing, the wheel-joint receptacle is a plain unthreaded cut, and the rib roots have no blends.

Fable 5.1 built a flat triangular plate with a single window, missing 13 of the 24 function claims. Like Astra it has no diagonal member, no auxiliary boss and no threads, and its plate and cross rib fill the fork gap between the pivot pads that the chassis bracket needs. It also turned the wheel-joint boss parallel to the pivot axis instead of perpendicular to it, and the part has 34% more volume than the source.

Inkling, the base model Euclid-1 was post-trained from, built a solid rectangular plate 42 m long, a thousand times the specified size. It has no branches, relief openings or blends; its two pivot pads stand on separate, non-coaxial axes with the fork gap between them filled in, and it missed all 24 function claims. Euclid-1, post-trained from this same base with RLFPC, scored 63.8 on the task, against 30.2 for Fable 5.1 and 27.7 for GPT-6 Astra.

05

What’s next

RLFPC got Euclid-1 here on a small fraction of the compute a frontier lab spends. That scales with compute, and so does what Euclid can design: from single parts to assemblies, to entire automobiles, aircraft and spacecraft.

Engineers stop moving geometry by hand and start directing agents that do - designing hardware at the speed of software.

The next industrial revolution is coming.