# Optimus Gen 2 · exterior illustration `optimus-gen-2.glb` is an **original Roboryn illustration**, authored with Codex, that interprets the exterior shown in Tesla's December 2023 Gen 2 introduction. It is deliberately approximate. It is **not Tesla CAD, an official Tesla mesh, a scan, a digital twin or a measured model**. The visual reference is Tesla's own [Optimus – Gen 2 film](https://www.youtube.com/watch?v=cpraXaw7dyc), published on 13 December 2023. The film was directly inspected at the frame positions retained in [manifest.json](manifest.json). That record includes original visual notes, the inspection date, recipe identity and output checksum. No Tesla video frame, texture or geometry is bundled. The original procedural recipe is the complete geometry source for this asset. ## Appearance and interpretation The model uses the film's qualitative exterior cues: a glossy dark visor and neck/shoulder region, a white continuous torso shell, white arm and leg shells, dark exposed joint regions, pale segmented hands and dark feet. It simplifies the shape, panel layout, finish and external details. No logo is included. All proportions are authored display choices. The numeric coordinates in the recipe and manifest are **uncalibrated display units**, not metres and not evidence of the robot's height, width, mass or clearance. Do not use this asset for dimension comparisons or import its coordinates into specification records. The dark shapes in visible gaps are simplified exterior surfaces. There are no modeled internal motors, batteries, electronics, sensors, wiring paths or verified complete assemblies. Separating the meshes shows the authored **visible groups**; it does not reveal a researched internal construction or a technical teardown. The rig's axes, names, pivot count and angle bounds exist only for interface animation. They are not Tesla degrees of freedom, safe operating limits, measured kinematics or evidence of real movement. The finger segments are separate visible geometry; individual fingers are not animated by this presentation rig. Any motion is a procedural illustration, not a motion capture, simulation or task demonstration. There are no embedded animation clips. ## Rights and source choice See [LICENSE](LICENSE) for the model's creation and usage notice. All shipped geometry and materials are authored in the recipe. No marketplace mesh is redistributed, and no Tesla CAD license or endorsement is claimed. A free CC BY model by Mat Rex on Sketchfab was considered. Its title did not identify a generation, and the ordinary download API required authentication (HTTP 401). It was not downloaded or used. Paid editorial models were also not used. No protected viewer resource was extracted and no purchase was made. ## Rebuild Use the existing repository installation with Node 24 and locked `three@0.186.0`. No Blender, Python, added package, network request or image generator is needed: ```sh node scripts/prepare-optimus-model.mjs ``` To verify an independent rebuild without replacing the current files: ```sh node scripts/prepare-optimus-model.mjs --output-dir /tmp/roboryn-optimus-rebuild cmp public/models/tesla-optimus-gen-2/optimus-gen-2.glb /tmp/roboryn-optimus-rebuild/optimus-gen-2.glb cmp public/models/tesla-optimus-gen-2/manifest.json /tmp/roboryn-optimus-rebuild/manifest.json ``` The recipe reads the existing manifest's source notes and inspection dates and preserves them. It writes the generated GLB and updates only recipe/output identity and output statistics in the resulting manifest. A rebuild is not a new reference-media capture and does not backdate knowledge or create data records. Change the explicit asset version when intentionally changing its interpretation. The geometry consists of original cross-section lofts, rounded boxes, cylinders and narrow visor edge strips. Indexed vertices are merged where their positions and normals match. Unused UV attributes are removed. No decimation, network texture, external buffer, decoder or runtime geometry generation is needed. The application serves the resulting GLB; it does not run this script on startup, build or deployment. ## Scene contract The root is `Roboryn_Optimus_Gen_2_Illustration`. It is already Y-up; front is +X, left is -Z and origin is the pelvis. The viewer should derive its framing and floor from the whole-model bounding box. Normalizing the view is not a scale calibration. Visible node names identify `head`, `torso`, `pelvis` and symmetric `left_` / `right_` shoulder, elbow, wrist, hand, hip, knee and ankle groups. The 15 authored revolute presentation pivots include the G1 viewer's motion names: - `left_hip_pitch_joint` / `right_hip_pitch_joint` - `left_knee_joint` / `right_knee_joint` - `left_ankle_pitch_joint` / `right_ankle_pitch_joint` - `left_shoulder_pitch_joint` / `right_shoulder_pitch_joint` - `left_elbow_joint` / `right_elbow_joint` Each joint has `extras.kind = "joint"`, `jointType = "revolute"`, `axis`, `lower` and `upper`. Cache the rest quaternion and apply the local-axis pose as `restQuaternion * axisAngleQuaternion`; restore the rest transform on reset. Each joint also carries an explicit `purpose` qualification. The remaining pivots are waist yaw, neck yaw/pitch and wrist roll. ## Exercised verification · 13 September 2026 - Independent offline regeneration on Node 24.7.0 / Three.js r186 produced byte-identical GLB **and manifest**. Both `cmp` checks passed. - Delivered GLB: **1,267,280 bytes**, **59,096 triangles**, **118 visible meshes**, **15 presentation pivots**, six materials, one embedded buffer, zero textures. - SHA-256: `bb4637e5786d02908fc599da5be91013930bf2ab6ac50180934b2ac2e95160c2`. - Three.js GLTFLoader loaded the actual file in isolated local headless Chromium with software WebGL. Front, side, rear and perspective renders were visually inspected. A floating visor stroke and shell/yoke overlap found in the first render were corrected, then those renders were repeated without console or page errors. This proves local loading and visible form, not manufacturer fidelity, mobile performance or a deployment. The GLB is presentation content. It does not create evidence-backed observations, assessments, preferred selections or editorial publication authority. Those use the project's separate research and human-review workflow.