Consulting
Simulation you can differentiate, all the way to the manufactured part.
We engineer physical devices — from scan to manufactured part — on our open-source CortenForge toolchain. It covers geometry, parametric design, meshing and fabrication, rigid- and soft-body physics, and sim-to-real calibration.
The spine
Most engagements start from the same problem. There is a real physical thing. You need a device that fits it, performs measurably, and can actually be made.
We keep that whole path in one differentiable representation.
One substrate
Scan geometry, device geometry, rigid dynamics, soft-tissue FEM, and the control policy all live in one representation — not five tools bolted together by file exchange.
A gradient that survives the path
The gradient runs from how the finished device performs on the real thing back to the geometry parameters and the controller. The hard part is contact — soft against rigid, with gradients running through it. We built it first, because it sets the ceiling for everything above it.
Tied to measurement
Every gradient inherits the model's error, so fidelity caps everything downstream. System identification and sim-to-real are first-class here, not a final polish step.
Most groups have either a differentiable simulator or a scan-to-fabrication pipeline. Having both, on one differentiable substrate, is the part you can't buy off the shelf.
What we do
Device co-design engagements
We design your device end to end — digital twin, design parameterization, optimization, manufacturing gate. Geometry and behavior are optimized together against a measurable physical objective, by gradients rather than parameter sweeps.
Scan → twin → fabrication
We ingest your scan, repair and weld it, and convert it to an SDF. From there we design the device as an implicit surface, with graded multi-material stacks and lattice or TPMS compliance fields. You get split molds, printability gating, and a generated fabrication procedure.
Sim-to-real calibration & validation
We instrument the real device, measure it, and fit the model back to the data — tractable precisely because the simulation is differentiable. You get the size of the gap, not an assumption about it.
Feasibility & architecture review
A short engagement answering three questions: can this be done differentiably, where does your accuracy cap out, and what would move it. You get a written assessment.
Where it applies
- Robotics and control
- Soft robotics and compliant actuators
- Biomechanics research
- Generative design and digital fabrication
- Custom-fit products
Assistive robotics is our flagship proving ground — the hardest exercise of the method, not the boundary of it.
Get in touch
Tell us about the part and the objective it has to hit.
viabalaena@proton.me