Blender Optics Simulator

Blender Optics Simulator

An optical bench you lay out in Blender, trace with real optics, and can hand to an AI agent.

Place lasers, mirrors, lenses, waveplates, gratings, crystals and detectors in 3-D. A live engine traces the beam through them — rays, Gaussian beams, polarization, dispersion — mounts everything on real opto-mechanics, and renders it in Cycles. The whole optical state is readable and writable over a localhost MCP bridge, so an agent works from measured geometry instead of guesses.

Sixteen seconds, rendered from a scene the add-on built. Full-quality MP4 .

Install

Requires Blender 4.2 LTS or newer (4.2+ / 5.x).

One-click, keeps itself updated. Open the [install page](https://emircbngl.github.io/blender-optics-simulator/) and drag the *“⤓ Drag this into Blender to install”* button onto an open Blender window. That installs the add-on and subscribes you to updates in one gesture. Turn on *Edit ▸ Preferences ▸ System ▸ Network ▸ Allow Online Access* first.

From a zip. Download optical_alignment_sim- .zip from Releases and use *Edit ▸ Preferences ▸ Add-ons ▸ Install from Disk…*. Updates then come from the add-on's own Updates panel.

Open the Optics tab in the 3-D viewport sidebar (press N).

Quick start

1. Setup ▸ Browse Examples… — pick *Michelson* to get a complete bench.
2. Simulate ▸ Trace — press Live. Move any part and the beam follows.
3. Setup ▸ Element — select an optic and change what it is: focal length, glass, coating, wavelength. Expand More for the rest.
4. Place ▸ Mount & Adjustment — put it on a real mount and turn the knobs, with − / + steps.
5. Inspect ▸ Optical Report — read power, polarization, path length and alignment error per detector; Align All walks the mounts until the beam lands where it should.

Headless, the same thing from Python:

import optics_api                                  # inside Blender: blender -b --python your.py

optics_api.build_example("michelson") # a full bench in one call optics_api.set_mount("MI_M_fixed", "KM100") # put a mirror on a kinematic mount optics_api.set_dof("MI_M_fixed", "TIP", steps=40) # turn a knob: the beam walks off optics_api.align_element("MI_M_fixed") # and back: 2.51 -> 0.0012 mrad print(optics_api.inspect_beam("MI_D")) # power, w(z), polarization, coherence

examples/ holds runnable scripts: michelson.py, mach_zehnder.py, agent_align.py, bell_entanglement.py, hong_ou_mandel.py.

What it does

• Traces a real beam. Ray paths plus Gaussian-beam propagation, Jones/Stokes polarization, Fresnel losses, dispersion, nonlinear conversion, interference and wavefront error. → what is modelled, and what is not
• 35 element types, 26 one-click benches. Mirrors through OPAs, prisms, gratings, crystals, spectrometers, adaptive optics. → element reference · full feature list
• Mounts on real hardware. Kinematic mounts, posts, cage systems, lens tubes and rails, with mechanical limits and collision checks. → hardware
• Measures like a bench. Detector power and polarization, beam profiles, path length, group delay and GDD, spectra, wavefront sensors. → dispersion, cylinders and spectra
• Aligns itself. Influence-matrix solvers walk the mounts: re-centre a beam, null a tilt, close an adaptive-optics loop.
• Renders what you built. Cycles/EEVEE with detailed optomechanics, animation renders, and SVG schematics. → how beams are drawn

A mirror knocked 2° out of alignment, then one align_element() call: pointing residual 7.02 → 0.0008 mrad. examples/agent_align.py reproduces it headlessly.

Drive it with an AI agent

The add-on exposes its full state as JSON over a localhost bridge and ships an MCP server, so an agent can read the bench and act on it:

get_state()   → every element's pose, ports, mount limits, beam path, detector readings

↓ decide set_param() · place_relative() · set_dof() · align_element() · ao_close_loop() · render() ↓ the beam re-traces get_state() → read the result, not a guess

Start it from Present ▸ Tools & Integration ▸ Start MCP Bridge. → agent guide · MCP server · tool list

Physics, honestly

Every push runs the physics: 341 textbook checks (Malus, Fresnel, Snell, the grating equation, Gaussian ABCD, Zernike orthonormality, energy conservation) plus a 618-check regression suite on both Blender 4.2 and 5.x. The core formulas were also verified against an external symbolic and numerical oracle; where that has not been done, the code and the docs say so. Every run builds the same benches in millimetre and metre scenes and requires the readouts to agree.

What is *not* claimed matters as much: this is a chief-ray engine with wave-optics overlays, not a full-wave solver. Thin elements carry no thickness, a grating has no blaze-efficiency model, and anything phenomenological says so where you read it. Model limits are written next to each feature.

→ scope and limits · element-by-element provenance · where the data comes from

Documentation

How to cite

A machine-readable CITATION.cff is included, so GitHub shows a Cite this repository button with ready-to-paste APA / BibTeX.

@software{cobanoglu_blender_optics_simulator,

author = {Çobanoğlu, Muhammet Emir}, title = {Blender Optics Simulator}, year = {2026}, version = {0.31.0}, doi = {10.5281/zenodo.20778997}, license = {GPL-3.0-or-later}, url = {https://github.com/emircbngl/blender-optics-simulator} }

The DOI above is the *concept* DOI and always resolves to the latest version; each release also mints its own version DOI (listed in CITATION.cff).

License & credits

GPL-3.0-or-later — see LICENSE. Vendor CAD and meshes are not included and remain the property of their owners; this project ships original metadata, procedural geometry and tooling.

Contributors

• Tengfei Ma — ShanghaiTech University · ORCID 0009-0008-4556-4682 · @Harca-Yita. Reports from a working optical bench in #1 drove wavelength-true beam colours and the shutter, the unit-scale work, shaped apertures, reflection at the coated mirror face, the Porro prism and polished mirror substrates, the OPA and group-delay work, and the groove-oriented grating, cylindrical lens and spectrum detector.

Built in the spirit of Bigweld's maxim from *Robots* (2005) — "See a need, fill a need."