Interface Guide

Willow 5 MOGEN: Platform Interface Guide

Welcome to the Cloud Oracle. The Willow 5 MOGEN interface is your central command for creating, testing, managing, and deploying deterministic spatial intelligence models. This guide breaks down the core features across the four primary workspaces: Learn, Vision, Models, and API.

1. The "Learn" Tab (Model Enrollment)

The Creator environment where you teach Willow new physical actions via Zero-Shot Enrollment.

Feature Breakdown:

  • Spatial Engine Mode (Master Control): The global toggle that dictates how the AI perceives space.
    • Universal (Scale-Invariant): The default mode. Evaluates structural form and normalizes all distances against the subject's torso length, allowing actions to match across subjects of any height.
    • Physical (Metric Space): Bypasses normalization to calculate exact physical distances in meters. Unlocks the Calibration Strategy menu:
      • Dynamic Reference Object: Select a standardized object (e.g., a Size 7 Basketball). The engine uses a 6D Kalman filter to track it (Node 76) and derive a strict metric scale.
      • Manual Triangulation (Pinhole): Bypasses object tracking. You provide the subject's exact height/weight globally to calculate scale via orthographic skew correction.
      • Synthetic Data (Pre-Scaled): Bypasses CV extraction entirely for pre-scaled .fbx, .json, or .npy uploads.

 

  • Configuration Engine (The 2x2 Grid):
    • Kinematic Zones (Wireframe): Clickable body segments (Head, Torso, Arms, Hands, Legs, Feet) that generate the 32-bit Zone Bitmask. Disabling zones here physically disables those sensors on the edge device to save CPU.
    • Signal Tolerance (DTW): A slider controlling the strictness of the Dynamic Time Warping algorithm. In Universal mode, this is a proportional threshold. In Physics mode, this becomes an absolute Spatial Tolerance Boundary measured in meters.
    • Tempo Control (Speed Gating): Toggles strict timing requirements. In Physics mode, this acts as a strict Velocity and Acceleration gate (e.g., ±20% variance).
    • Continuous Action Setup (Overlap Tolerance): Controls Non-Maximum Suppression (NMS). Defaults to 25% to prevent "machine-gun" duplicate triggers during rapid, consecutive repetitions.

 

  • Research Mode (DSP Bypass): An advanced toggle that disables the Kalman Filter and Savitzky-Golay interpolation. Used strictly by data scientists and roboticists who require the raw, unadulterated ground-truth sensor data for high-fidelity physics calculations.

 

  • Upload Dropzone: Accepts standard RGB Video (.mp4, .mov) for empirical models, or pre-processed 3D virtual data (.fbx, .json, .npy) for synthetic models.

 

  • Precision Trimmer & Auto-Sync: When uploading videos, this modal allows you to surgically isolate the action data before processing. Features include:
    • Set Start / End (In/Out Points): Locks the exact boundaries of the kinetic action.
    • Extrinsics Row (Angle, Height, Distance): Define the specific Camera Angle. Note: If you selected "Manual Triangulation" in the Master Control, dynamic Camera Height (m) and Camera Distance (m) inputs will appear here to calculate pinhole depth for this specific video.
    • Auto-Sync (Clap): If uploading multiple camera angles of the same action, this button uses audio waveform peaks to automatically align the video feeds down to the millisecond before 3D fusion.
    • Link Synced Clips: A toggle that locks multi-angle videos together after syncing, ensuring that adjusting the trim on one camera angle perfectly mirrors the trim across all other angles.

2. The "Vision" Tab (Action Detection Engine)

The Vision Lab. A high-performance sandbox to verify model resilience, tune tolerances, and simulate edge environments.

Feature Breakdown:

  • Sub-Tabs (New Analysis vs. History): Toggle between launching a new detection scan and reviewing the historical archive of past detection reports.

 

  • Target Configuration & Queue: Allows you to load one or multiple enrolled models simultaneously. If a Physics model is loaded, the engine automatically primes YOLO to track the required reference object.

 

  • Processing Strategy & FPS Configuration:
    • Production (Scan & Seek): Simulates a real-world edge device. Allows you to define a Scan FPS (a low-framerate wide-net scan) followed by a Seek FPS (a high-framerate targeted extraction for absolute verification).
    • Research (Direct Pass): Forces the engine to process the entire video at maximum fidelity (RAM intensive; used for deep data-science validation).

 

  • Tactical Bounding Box Renderer (Playback Lightbox):
    • Action Timeline Track: Color-coded blips on the timeline indicating exactly where the engine triggered a match.
    • Live HUD: Draws strict bounding boxes around the subject during the triggered action, displaying the matched Model Name and the exact Confidence Percentage in real-time.

3. The "Models" Tab (Library & Management)

Your central repository of enrolled kinetic signatures, offering deep inspection and JIT (Just-In-Time) exporting.

Feature Breakdown:

  • Sub-Tabs (All Models vs. Stages): Toggle between your full model library and Deployment Stages. Stages allow you to group specific models together into deployment environments (e.g., 'Robotic Arm Alpha'). Your edge applications can poll a specific Stage API endpoint to dynamically sync only the models required for that application.

 

  • Model List & Consensus Matrix: Displays all enrolled models and a visual matrix showing which camera angles were successfully fused. Includes prominent UNIVERSAL and PHYSICS badges to instantly identify the underlying mathematical engine used.

 

  • Inspect (Model Intelligence): A drawer that reveals the baked-in configuration (Spatial Mode, Signal Processing status, Zones, Speed Gate). For Physics models, this includes a Calibration Diagnostics readout displaying the Calculated Anatomical Height and Metric Scalar to mathematically prove perspective correction.

 

  • Calibrate (Firmware Lock): Allows engineers to tweak the Firmware Calibration Constants (DTW Sensitivity, Tempo, Overlap) after the model is enrolled. These updates are locked directly into the database and will be baked into the header of the next export.

 

  • View 3D (Ghost Lightbox): An interactive Three.js environment that visualizes the raw math of your model.
    • Integrity Scorecard: A detailed readout showing the visibility confidence percentage of every body zone. Low scores indicate high occlusion during enrollment.
    • Metric Space Visualization: If viewing a Physics model, the UI renders an active metric badge, a 0.5m physical floor grid, a 2.0m holographic ruler, and a glowing orange tracking sphere representing Node 76 (the physical object interacting with the human).

 

  • Export (IRaaS Signature): The Just-In-Time compiler gateway. Select your target format:
    • .int8 (Quantized binary for Edge/Chip)
    • .onnx (AI/ML Pipeline Graph)
    • .h (Embedded Firmware)
    • .json / .csv (Data Science)

4. The "API" Tab (Developer Integration)

The programmatic gateway for B2B integrators to automate the Cloud Oracle and provision edge fleets.

Feature Breakdown:

  • Connection Credentials: Securely displays your x-customer-id and the SDK Unified API URL required to route requests to your specific cloud partition. Includes "Reveal/Hide" mask toggles to prevent shoulder-surfing of sensitive tokens.

 

  • Active API Keys Table: Manage your programmatic access. For security, accounts are limited to 3 active keys at any given time.

 

  • Generate / Revoke Capabilities: Instantly generate new secure tokens or permanently revoke compromised keys. Revoking a key instantly severs access for any edge SDK or automated pipeline utilizing it.