Blender Optics Simulator

An optical bench an AI agent can build, inspect, align, and render — over MCP.

"See a need, fill a need." — Bigweld, Robots (2005)

60-second tour — Lay out lasers, mirrors, beamsplitters, lenses, waveplates, gratings and detectors in Blender; a live beam engine traces them (ray + Gaussian-q ABCD + Jones/Stokes polarization + wave-optics overlays), all physics-verified against textbook answers in CI. Everything mounts on real opto-mechanics and renders in Cycles. The whole optical state is exposed over a localhost MCP bridge, so an AI agent reads ground-truth geometry and beam data and drives the bench — aligning, closing an AO loop, nulling fringes. Install: drag the one-click link from the releases page into Blender 4.2+ (it also subscribes you to updates — see Install & stay updated).

Blender becomes a physics-checked optical bench: lasers, mirrors, beamsplitters, lenses, waveplates, polarizers, gratings, deformable mirrors, detectors — laid out in 3-D, traced by a live beam engine, mounted on real opto-mechanics, and rendered in Cycles. The twist: the entire optical state is exposed as JSON over a localhost MCP bridge, so an AI agent (e.g. Claude) doesn't guess geometry — it reads every element's pose, port normals, beam path, mount limits, and wavefront error as ground truth, then drives the bench toward design intent and watches the beam update live.

This is the one place Blender × optics × MCP/AI-agents actually overlap. Most Blender-MCP work is geometry and animation. This is a physics-grounded digital twin of a real optical table that an agent can see and close the loop on.

The core loop

An agent doesn't fly blind — it has the optical truth and iterates against it:

get_state()  →  read every element's world pose, port positions+normals, mount DOFs,
                 traced beam path, detector readings, wavefront error  (full JSON)
   ↓ decide
action()     →  align a mirror · scan OPD · place a part on a grid hole · close the AO loop
   ↓
beam re-traces live  →  next iteration

get_state() returns the whole bench as JSON — elements, sources, detectors, beam path, bench grid, cages/tubes/rails, and mechanical-limit warnings — so the agent always knows where things are and where the light goes. 30 functions are exposed over MCP; the bridge whitelists only the curated optics_api facade.

Related MCP server: Blender MCP

Drive it with an AI agent (MCP)

1. In Blender: Optics ▸ Simulation ▸ Start MCP Bridge (exposes every public optics_api function on 127.0.0.1:9765).

2. Run the bridge server in mcp/ and wire it into your MCP client.

Then the agent can READ the bench, WRITE changes, and COMPOSE subsystems:

import optics_api

# COMPOSE — bootstrap a canonical bench in one call
optics_api.build_example("michelson")        # full setup + live beam

# READ — the agent's numeric eyes: ground truth, not guesses
state = optics_api.get_state()               # poses, port normals, beam path, mounts, detectors
optics_api.inspect_beam("MI_BS")             # power, w, R(z), M², divergence, polarization, coherence here
optics_api.inspect_element("MI_BS")          # what this optic DOES + its live in/out by kind, throughput
optics_api.beam_profile("MI_D")              # Gaussian w(z), waist, aperture clipping
optics_api.detect_phenomena()                # flags interference / off-axis-hologram conditions the trace meets
optics_api.propose_corrections()             # advisory fixes you JUDGE (refuse / partial / accept)

# WRITE — act, then the beam re-traces
optics_api.align_all()                       # auto-tune every kinematic knob toward its target
optics_api.set_param("MI_BS", "split_ratio", 0.5)
optics_api.place_on_grid("MI_M_fixed", 6, 2) # slide a part to a real breadboard hole
optics_api.scan(kind="STAGE", lo=0, hi=2e-3, steps=120, element="MI_M_stage")  # interferogram PNG+CSV

# Adaptive optics: sense the wavefront, then close the loop
optics_api.build_example("adaptive_optics")
optics_api.ao_measure("AO_WFS")                                 # {zernike:[...], rms: 0.559} (incl. beam defocus)
optics_api.pyramid_wfs("AO_WFS")                                # same wavefront read as a SLOPE sensor
optics_api.ao_close_loop("AO_WFS", "AO_DM", gain=0.5, iters=15) # rms 0.559 → ~0

# RENDER — photorealistic Cycles, or a turntable movie
optics_api.bake_beams()
optics_api.render(preset="final", camera="HERO", filepath="/tmp/michelson.png")

The full surface (each also an MCP tool): capabilities, get_state, diagnose, propose_corrections, detect_phenomena, inspect_beam, inspect_element, beam_profile, sensor_capture, ao_measure, get_wavefront, pyramid_wfs, zonal_render, coupling_efficiency, check_mechanics, build_example, trace_beam, tag_element, add_component, swap_part, place_relative, set_mount, set_param, align_element, align_all, auto_align, tilt_null, design_telescope, design_4f, mode_match, scan, render, render_sequence, bake_beams, clear_beams, export_svg, dress_bench, set_grid, place_on_grid, make_cage, make_tube, make_rail, place_on_rail, ao_command, ao_close_loop. Call capabilities() first — it returns a self-describing manifest (read mcp/AGENT_GUIDE.md for the conventions and the refuse / partial / accept judgement model).

Five repeatable /optics-* skills ship in .claude/skills/ (build · align · inspect · correct · sensor-render) — model-invoked workflows that sequence these tools and carry the disciplines (inspect-first, byte-identical, advisory-corrections-you-judge).

For headless pipelines, optics_api is importable directly inside Blender (blender --background --python your_script.py).

What you can do

• Design a real bench — 26 optical element types on real opto-mechanics: tapped-hole breadboards (metric 25 mm / imperial 1″), 9 real mount presets (KM100/KS1/POLARIS-class kinematics, GM100 gimbal, RSP1 rotation, TRF90 flip, VC1 V-clamp — real DOF ranges, detents and clamping), Ø12.7 mm posts + post-holders, 16/30/60 mm cage systems, SM05/SM1/SM2 lens tubes, and RLA dovetail + X95 structural rails with carriers — all original GPL-clean procedural geometry to true functional dimensions, structurally checked in CI.

• Simulate the light — a live beam tracer with analytic overlays: polarization (Jones/Stokes, plus a per-pixel DoFP polarization camera), interference & fringe visibility, Gaussian-beam propagation (ABCD w(z), ROC, Gouy, M²), modal and pyramid wavefront sensing + closed-loop adaptive optics, wavelength-selective materials (interference/colored-glass filters, soft-edge dichroics), nonlinear conversion (SHG/SPDC), and acousto-optic Bragg deflection.

• Let an AI drive it — the agent READS numeric ground truth (inspect_beam / inspect_element / get_state), is told the phenomena the bench meets (detect_phenomena: interference, off-axis hologram) and the advisory corrections it should weigh against your intent (propose_corrections: refuse / partial / accept), and follows repeatable /optics-* skills — never guessing, every formula oracle-verified.

• Align — kinematic auto-alignment that drives only the mount knobs to < 1 mrad and reports "move the post" when a target is out of reach; a geometric validator flags collisions and posts in the beam.

• Render — bake beams to glowing emission tubes, one-click EEVEE preview / Cycles final with glass materials + studio lighting, camera presets, transparent-PNG figures, turntable movies, and publication-ready SVG schematics.

26 one-click example scenes

Each is a single build_example(kind) — built from portable, mesh-free generic components, fully agent-drivable:

…plus ten more: prism (dispersing-prism spectrometer, white light fanned by n(λ)), beam_router (rhomboid parallel-displace + tilt-invariant penta steering), beam_profiler (knife-edge erf scan), back_reflection (a window's parasitic Fresnel retro-ghost), green_doubler (1064→532 nm SHG), spdc_source (degenerate 810 nm signal+idler pairs), quad_tracker (4-quadrant position error), circulator (non-reciprocal 3-port fiber router), and the surface_figure_native / surface_figure_diverging WFS variants.

Gallery

Everything below is produced by the add-on itself — a viewport beam trace baked to emission tubes, the Scan + Plot operator, the Detector Fringe Image, and the live sensor window exported to PNG.

New — catalog-true cage hardware + a structural-support gate

Cage assemblies now build from real catalog parts: rods snap to the vendors' fixed ER/SR lengths (a 130 mm train picks the ER6 · 152.4 mm rod and shows the true symmetric overhang; cage_info() reports the part), a single-member cage stands rodless on its plate, the assembly is post-mounted through one plate's bottom tap, plate bores close onto each optic's actual mesh, and lens tubes carry real retaining rings. Behind it sits a new structural-support gate: support_scan() measures the mesh-level gap between every optic and the hardware meant to hold it, and validate()/diagnose() flag anything unsupported — so floating hardware cannot ship silently, in any scene.

New in v0.25.0 — expanded opto-mechanics: 9 mount presets, X95 rails, structural QA in CI

4 new mount presets — RSP1 rotation, GM100 gimbal, TRF90 flip with real 0°/90° detent semantics, VC1 V-clamp for cylindrical bodies — plus the X95 structural rail family next to the RLA dovetail. The whole opto-mechanical catalog was rebuilt to true functional geometry: bases, hinges and clamps seat correctly for any beam direction (including folded and vertical paths), preset apertures use the vendors' real dimensions, and every optic sits in genuine contact with its hardware. A render-free structural check now runs in CI on every push (tests/test_mesh_health.py: mesh integrity per element, part-contact analysis per mount assembly), so the catalog cannot silently regress.

New in v0.24.0 — laser speckle + the coffee-cup caustic (phenomena closed out)

Two off-trace phenomenon producers, each derived from first principles and checked against its exact textbook result. speckle_pattern() — a random-phase diffuser propagated to a screen gives fully-developed laser speckle: contrast σ_I/⟨I⟩ → 1, a negative-exponential intensity PDF, mean grain ~ λz/D, and 1/√N suppression when you average N frames (contrast + averaging law physics_verify ok=true). caustic_pattern() — parallel rays reflecting off a circular mirror's concave wall pile up on a nephroid (the coffee-cup caustic), the ray-density lighting up the analytic envelope ~20× with its cusp exactly at the mirror focus R/2 (reflection geometry + cusp physics_verify ok=true). This closes the phenomenon-emergence catalog (hologram · interferogram · Fabry–Pérot · Talbot · Newton's-rings · speckle · caustic). Also new: diagnose() warns when an optic is knocked off the beam (optic_bypassed) — a mid-path lens/mirror/BS the beam no longer reaches now raises a WARN with a re-centre fix, closing a gap a render surfaced.

New in v0.23.0 — opto-mechanical hardware rides its optic (rigid assemblies)

Dressed hardware (post, holder, mount) used to be built as standalone objects at fixed world coordinates, so grabbing an optic — e.g. a lens in the newton_rings bench — left its mount behind. Now each cluster's hardware is rigid-parented to its optic via matrix_parent_inverse, so the whole assembly moves as one body: the parenting causes no visible jump, leaves the optic's matrix_world untouched, and the live trace stays byte-identical (regression 392/392). Also new: inspect_all() (a per-element dashboard) and export_report() (a self-contained HTML spec-sheet of the whole bench), the always-visible in-add-on update button, four more glasses + CLBO/AgGaS2 + the As2S3/AgCl/ZnS infrared materials (sourced, with a new docs/DATASOURCES.md provenance doc), and a distinct translation-stage mount.

New in v0.22.0 — biaxial nonlinear crystals (KTP + LBO) + Newton's-rings 2D

The Sellmeier-derived SHG phase matching covered only uniaxial crystals; this adds the two workhorse biaxial doublers, KTP and LBO, with all three principal-axis Sellmeier sourced from refractiveindex.info. In the XY principal plane the biaxial problem reduces to an effective-uniaxial one, and from the sourced coefficients the solver reproduces the textbook Nd:YAG green-doubler cuts — KTP Type-II φ=23.58° (lit 23.5°), LBO Type-I φ=11.76° (lit 11.6°) — plus their spatial walk-off (KTP ~4 mrad, LBO ~7 mrad). Off-trace calculators; the live trace stays byte-identical. (The in-plane index order was caught and corrected via physics_verify before coding — the polarization is perpendicular to the wavevector, so 1/n² = sin²φ/n_x² + cos²φ/n_y².)

Also new: newton_rings produces the 2-D Newton's-rings reflected pattern of a plano-convex surface on a flat — I(r)=sin²(πr²/λR), a central dark spot and dark rings falling exactly on r_m=√(mλR) (docs/img/newton-rings-demo.png).

New in v0.21.0 — birefringence: o/e double refraction + a χ² coupled-wave solver

The two biggest architectural items on the backlog — both in the live tracer, yet byte-identical (each behind a per-element opt-in flag, default off). oe_split gives a uniaxial crystal true ordinary/ extraordinary spatial double refraction: one beam in, two orthogonally-polarized beams out, the extraordinary one walked off by L·tan(ρ) (the calcite double image). use_chi2_solver derives a nonlinear crystal's SHG efficiency from the full Manley-Rowe coupled-wave ODE — pump depletion + phase mismatch, RK4-integrated, reproducing tanh²(√η_lin) at phase match and η_lin·sinc² undepleted (both verified to ~1e-12), with the walk-off derived from the index ellipsoid. The tracer already branches rays (beamsplitter, Wollaston) and emits walk-off SHG children, so each is the same split-two pattern — not a new engine.

New in v0.20.0 — wave optics: aberrated PSF, phase retrieval & cavity modes

aberrated_psf — the forward problem (wavefront → image). wave_psf could only add Z4 defocus; this aberrates the diffraction PSF with any Zernike mode (defocus, astigmatism, coma, trefoil, spherical) at a chosen RMS, or a full Noll-indexed wavefront vector. The Strehl ratio collapses the same way for every mode — it depends only on the RMS wavefront error (the Maréchal approximation exp(−(2π·rms)²): 0.05 waves → 0.906, 0.10 waves → 0.674) — but the PSF morphology is the mode's fingerprint: defocus blurs symmetrically, astigmatism elongates, coma flares one-sided, spherical haloes.

fienup_phase_retrieval — the inverse problem (the genuine phase problem). A detector records only the diffraction intensity |FFT(object)|²; the phase is lost. Given that magnitude plus a real-space support mask, Fienup's Hybrid-Input-Output algorithm reconstructs the hidden object (correlation 0.996, up to the inherent translation + twin ambiguity). Where v0.19.0's Gerchberg-Saxton knows the amplitude in both planes (CGH design), here the object is unknown — only its support is; the HIO feedback escapes the stagnation pure error-reduction falls into.

tem_mode — the laser cavity transverse eigenmodes a real resonator emits. family="HG" gives the Hermite-Gaussian TEM_mn (rectangular, (m+1)(n+1) lobes, an orthonormal set); family="LG" gives the Laguerre-Gaussian donut LG_{p,l} with an on-axis null and an exp(i·l·φ) phase vortex carrying orbital angular momentum l·ħ. All three additions are off-trace; the geometric trace stays byte-identical. (This release also fixes a latent NameError that would have crashed the gerchberg_saxton / spatial_filter / propagate_chain MCP tools when called by an agent, and adds an end-to-end guard so it can't recur.)

New in v0.19.0 — Fourier optics: phase retrieval + spatial filtering

A Fourier-optics pair on the verified FFT: gerchberg_saxton (the iterative Fourier-transform algorithm) designs a source-plane phase mask — a computer-generated hologram — whose far-field matches a target, with the monotone-error convergence guarantee; spatial_filter (the Abbe-Porter experiment) applies a Fourier-plane mask (lowpass / highpass / Zernike phase contrast that makes a pure-phase object visible). Both off-trace; the geometric trace stays byte-identical.

New in v0.18.0 — multi-plane field-propagation chain (propagate_chain)

propagate_field runs the angular-spectrum propagator once; the new propagate_chain marches a complex field through a sequence of planes — a POPPY-style multi-plane OpticalSystem built from prop / aperture / lens steps (e.g. aperture→lens(f)→prop(f) focuses at z=f). Closes the optics-textbook catalog's principal remaining gap (single plane→plane is now a chainable auto-pipeline). Validated against propagator additivity, focus-at-f, and the free Gaussian w(z). Off-trace; the geometric trace stays byte-identical. (Near-field layer — a tight focus or Fraunhofer far field is better via direct FFT.)

New in v0.17.0 — dichroic AOI edge + Sellmeier transparency window

Two physical-honesty refinements: a thin-film dichroic mirror's edge now blue-shifts with the angle of incidence (a 45° dichroic sits ~8% to the blue of its normal-incidence spec — the fluorescence-microscopy gotcha), consistent with the interference filters; and sellmeier_in_range flags when a refractive-index query falls outside a glass's published fit window (an extrapolation, not a measured value). Both keep the geometric trace byte-identical.

New in v0.16.0 — Fabry-Pérot + Talbot emergence (phenomenon catalog complete)

produce_phenomenon now covers all four phenomena: detect_phenomena also flags a Fabry-Pérot cavity and a Talbot grating, and produce_phenomenon emerges the Airy transmission resonance (finesse, FSR, contrast) and the Talbot self-imaging (z_T = 2 d²/λ) — behind the same advisory accept-gate. Validation now 163 oracle checks.

New in v0.15.0 — phenomenon emergence (conditions met → the phenomenon is produced)

detect_phenomena flags when an optical phenomenon's conditions are met; the new produce_phenomenon now actually produces it — synthesizing the two-beam interferogram and recording + reconstructing an off-axis hologram (carrier Λ = λ/(2 sin θ/2) → 3.63 µm, the object beam's angle recovered from the carrier peak). It stays advisory + intent-judged like propose_corrections: a dry-run with an intent caveat first ("the camera may be a power meter, not a hologram plate"), then accept=True emerges the pattern — never silently. Off-trace; each result self-checks against a textbook oracle.

New in v0.14.0 — uniaxial crystal optics + Mueller calculus

Closed-form anisotropic-ray numbers on the existing ordinary/extraordinary index catalog: birefringent walk-off angle (calcite at 45° → 6.224°), the index ellipsoid n_e(θ), true zero-order waveplate thickness (quartz QWP = 16.19 µm), and the Type-I SHG phase-matching angle (BBO 1064→532 nm → 22.78°). Plus a 4×4 Mueller calculus (M_mueller_polarizer/M_mueller_retarder/stokes_through) for partially-polarized light, validated against py-pol / SymPy / Malus, and end-to-end CI coverage of the waveplate crystal-birefringence dispersion branch. Validation now 152 oracle checks.

New in v0.13.0 — diffraction & focusing examples + a verified FDTD bridge

Single/double/N-slit Fraunhofer diffraction, the Talbot self-imaging carpet, and a Fresnel diffraction & focusing batch (circular-aperture zones, knife-edge, Fresnel zone-plate, thin-lens focus, Fabry–Pérot Airy minimum, Newton's rings) — each reproduced by the field engine and pinned to its textbook closed form (validation now 136 oracle checks). And the FDTD bridge's Meep API is now verified against a real Meep 1.33.0 (12/12 vs exact analytic oracles; tools/verify_fdtd_meep.py), with three real setup bugs fixed. Core trace byte-identical to v0.12.0.

New in v0.12.0 — the physical-optics field engine (opt-in, off-trace)

A complete sampled-field layer on top of the geometric + Gaussian-q core. Every panel is computed on demand and never touches the live trace (the byte-identical regression is preserved); every formula is machine-verified against the physicist Docker oracle (the validation suite now runs 136 checks).

Plus fdtd_derive_property (orchestrates Meep/Tidy3D for a grating/coating/metasurface, closed-form fallback when absent), tolerance_scan (pose-only Monte-Carlo alignment yield), and the /optics-scope skill that answers "can it simulate X?" honestly — by tier, with the verified case or the external tool to use.

Wavefront sensing — modal and zonal, honest about the difference. The same surface figure reads three ways through three beams; the dense zonal map keeps high-spatial-frequency relief the 15-mode modal fit smooths away; recognizable objects (a chess knight, a die) make the map legible.

One-click Dress Bench seats every optic at one beam height on a real tapped-hole breadboard (metric 25 mm or imperial 1″) with mount-type-correct hardware — kinematic mirror mounts (KM100/KS1-style), a cube beamsplitter mount, threaded lens cells, rotation, gimbal, flip and V-clamp mounts — on Ø12.7 mm posts in post-holders. Decoration only (never traced), so a Cycles render reads like a real table. It's also a data model: get_state() reports the grid, beam height, vertical post chain, and which hole/system each optic uses — so an agent knows exactly where things go.

Every mounting system is original GPL-clean procedural geometry built to real functional dimensions (Thorlabs SM threads, ER cage rods, RLA rail) — vendor CAD is never bundled. Grouping optics into a cage/tube/rail never moves them, so the beam path is byte-identical.

A complete infinity-corrected microscope: a Köhler lamp + condenser image the illumination onto the sample, the objective collimates into the infinity space, and a tube lens forms the image at the camera. The objective is a real element — f_obj = f_tube / M (oracle-verified: 10× on a 200 mm tube lens → 20 mm), with a numerical aperture, working distance, and DIN colour ring. Every lens applies true Gaussian-beam (ABCD) focusing, so the beam genuinely converges.

An acousto-optic modulator / Bragg cell — a clear TeO₂ crystal with a piezo transducer launching a travelling acoustic grating. The beam splits into the undiffracted 0th order and a frequency-shifted +1 order deflected by θ = λ·f_a/v_s — the oracle-verified Bragg relation (shear-mode, v_s ≈ 650 m/s, f_a = 200 MHz → a visible ≈ 11°). Deflection angle, grating period, and frequency shift are all physics-checked in the sandbox.

Components & physics. Real ray-bending elements on real opto-mechanics — every behaviour from element properties, every formula oracle-verified.

Methods — physical & digital

Two halves: the physics the simulator models, and the engineering that keeps it honest.

Physical methods (the optics, all in a dependency-free physics.py):

• Geometric chief-ray tracing through sequential elements — Snell refraction, vectorial Fresnel, multi-bounce splits.

• Polarization — Jones vectors/matrices, Stokes / DOP, Malus, exact 3-D Fresnel s/p phase, and a per-pixel DoFP polarization camera.

• Gaussian beams — the complex q-parameter through ABCD systems: w(z), R(z), Gouy phase, beam quality M², mode-matching.

• Wavefront sensing — modal 15-Noll-Zernike and pyramid (slope) reads, closed-loop adaptive optics, and dense zonal surface-figure maps (W = 2·cos²θ·Δdepth).

• Wavelength — Sellmeier dispersion n(λ), soft-edge dichroics (R(λ) + T(λ) = 1), interference / colored-glass filters, the grating equation.

• Interference & phenomena — coherent recombination + fringe visibility, off-axis-hologram carrier spacing Λ = λ/(2 sin θ/2), nonlinear χ² conversion (SHG / SPDC), acousto-optic Bragg deflection.

Digital methods (the engineering that makes it trustworthy):

• Property-driven & byte-identical — every behaviour is element properties × matrix_world; the mesh is cosmetic and the trace is deterministic until you act (the architecture below).

• Oracle-verified — every core formula is machine-checked against an external symbolic + numeric oracle (units, identities, limits, known-input → known-answer) before it ships: verified, not "dimensionally plausible."

• A 357-check regression harness runs headless in Blender on every push (CI-green), and opt-in features stay trace-byte-identical.

• AI-drivable over MCP — the agent loop read → judge → act → re-trace, advisory corrections you judge (refuse / partial / accept), and five /optics-* skills.

• Reproducible figures — the analysis figures are composed by tests/_plot_*.py from dumped trace data through a shared figure schema (figstyle) that guarantees the colorbar and captions never overlap or clip, and is re-themeable without touching the layout.

Physics & scope (honest)

The live trace is a single-ray (chief-ray) geometric tracer (tracer.py) with analytic physics overlays — Jones/Stokes polarization, Gaussian-beam ABCD propagation, and analytic interference / fringe models, all in a dependency-free physics.py. The live trace itself is not a wave-optics solver — it stays instant and byte-identical. On top of it sits an opt-in, on-demand physical-optics field engine (new in v0.12.0) — the real diffraction PSF, sampled-field propagation, turbulence, nonlinear pulses, Monte-Carlo transport, FDTD orchestration — that you call explicitly and that never touches the live trace, so the byte-identical regression is preserved. It is not a lens-design optimizer.

But what it does model, it models honestly. The core formulas below are checked against an external symbolic + numerical oracle (the physicist verifier: units, symbolic identities, limits, and known-input → known-answer) — so they are verified, not merely "dimensionally plausible." The broader element-variant catalog is physically modeled but not each independently oracle-checked — see docs/OPTICAL_ELEMENTS.md for per-element provenance.

Two additions stay deliberately honest about their tier: detect_phenomena flags the conditions a phenomenon needs (it does not produce a diffractive image), and the pyramid WFS reports the wavefront gradient a pyramid integrates — a geometric slope, not a diffractive 4-pupil image (the chief-ray tracer does not propagate the pupil-plane field). Both are labelled Tier-1 in the code and docs.

Run the self-test with a bare interpreter (no Blender needed):

python3 optical_alignment_sim/physics.py   # -> PHYSICS SELFTEST PASSED

Catalog & meshes. A built-in, IP-clean component library — our own, built to real market norms: 38 catalog entries whose dimensions, specs, and mount conventions are modeled on industry-standard parts (the kind you'd source from Thorlabs, Edmund, Newport, …), addressable by SKU, plus 118 documented real-world variants in docs/OPTICAL_ELEMENTS.md. No vendor CAD is bundled — vendor 3-D models are the vendors' IP. The library ships only original metadata (element type, port geometry, mount parameters) and correct generic mesh-free geometry at true functional dimensions, so every entry is usable out of the box. Import your own CAD (STL/OBJ natively; STEP/IGES via FreeCAD) to drop in an exact part.

Install

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

One-click — recommended, auto-updates

Open the install page and drag the “⤓ Drag this into Blender to install” button onto an open Blender window. Blender installs the add-on and subscribes you to update notifications in one gesture — no URL to type. (Make sure Edit ▸ Preferences ▸ System ▸ Network ▸ Allow Online Access is on.)

Every future release then appears inside Blender — a status-bar badge, then Preferences ▸ Get Extensions ▸ Install Available Updates. You never download a zip by hand again.

Then open the Optics tab in the 3D viewport sidebar (press N).

From a GitHub zip — updates from inside the add-on

Download optical_alignment_sim-<version>.zip from the Releases page and use Edit ▸ Preferences ▸ Add-ons ▸ Install from Disk…. (Or build from source: blender --command extension build --source-dir optical_alignment_sim --output-dir .)

Blender's own "Install from Disk" copies live in a local repository it never auto-syncs — but you still get updates: the add-on's Updates panel (in the Optics sidebar) has an always-visible "Up to date · Check" control. Pressing Check registers the project's update channel for you and pulls the newest version through Blender's native extension system — no manual re-download, no re-install. (Needs Allow Online Access on.)

Install & stay updated — every path is covered

Either way, the check runs at most once a day and only when Allow Online Access is enabled; nothing phones home otherwise.

Quick start

From the UI

• Examples ▸ pick Michelson (or any of the 26) to spawn a full setup with the live beam overlay.

• Element — select an object, Tag as Optical Element, Auto-Detect Ports; per-type parameters appear (source polarization, waveplate angle, lens focal length, …).

• Mount & Adjustment — Apply Mount Preset (e.g. KM100CP/M), Set Coarse Pose, drive the tip/tilt knobs.

• Simulation — toggle Live simulation; the beam updates as you move parts. Start MCP Bridge to let an external agent drive the scene.

• Alignment Report — Update Report / Align / Align All; detectors show measured power, polarization, and fringe visibility.

• Adaptive Optics — Run AO Loop to sense a wavefront and drive a deformable mirror flat.

• Render — pick a camera + Background preset, then EEVEE Preview / Cycles Final (toggle Realistic optics for glass + studio lighting); Dress Bench for the full table; Export SVG Schematic for a publication-ready vector figure.

Headless / standalone scripts

examples/ contains builders that run with or without the add-on installed:

blender --background --python examples/agent_align.py    # the agent-align demo, headless
blender --background --python examples/mach_zehnder.py

How this compares

A layout, alignment, visualization, and agent-control tool — not a lens-design optimizer or a wave-optics solver. It sits in a niche the standard tools don't cover:

The unfair advantage is the last column. Need a diffraction PSF or a tolerancing run? Reach for Zemax/POPPY/prysm. Want physically precise lens/mirror meshes to render? OptiCore is excellent at exactly that. This tool is the other half: it doesn't just place accurate geometry — it runs a live beam through it (ray + Gaussian-q + Jones/Stokes polarization), auto-aligns the bench with influence-matrix solvers, flags problems with diagnose(), and exposes the whole state so an AI agent can drive it over MCP — the Blender × optics × MCP intersection almost nobody occupies.

How to cite

If you use Blender Optics Simulator in academic work, please cite it. 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 = {0.27.0},
  doi     = {10.5281/zenodo.20778997},
  license = {GPL-3.0-or-later},
  url     = {https://github.com/emircbngl/blender-optics-simulator}
}

This software is archived on Zenodo with a citable DOI: 10.5281/zenodo.20778997 — the concept DOI, which always resolves to the latest version. Each release also mints its own version DOI (see CITATION.cff).

License & credits

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

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