Upload README.md with huggingface_hub

README.md

@@ -0,0 +1,49 @@
+# Model Card: 5-Task Robot Classifier
+
+## Model
+- Name: `best_licensed.pt`
+- Architecture: `MobileNetV3-Small` classifier
+- Tasks: `cleaning`, `cooking`, `dishwashing`, `laundry`, `organizing`
+
+## Intended Use
+- Classify household task from RGB frame for non-safety-critical robot workflows.
+
+## Out of Scope
+- Safety-critical decisions
+- Medical/legal/security use
+- Tasks outside the 5-class scope
+
+## Training Data
+- Source: Licensed/owned dataset only
+- License ledger: `data_licenses.csv`
+- Dataset version: `licensed_dataset_v1` (fill exact hash/version)
+
+## Evaluation Protocol
+- External set: `isaac_eval/labels.csv` + frames
+- Command:
+```bash
+python isaac_eval.py --checkpoint best_licensed.pt --labels-csv isaac_eval/labels.csv --images-root isaac_eval
+```
+
+## Reported Metrics (fill before publish)
+- Overall accuracy: `__%`
+- Cleaning: `__%`
+- Cooking: `__%`
+- Dishwashing: `__%`
+- Laundry: `__%`
+- Organizing: `__%`
+
+## Known Failure Modes
+- Visual confusion between cleaning vs organizing in cluttered environments
+- Domain shift from training camera to deployment camera
+- Low light / motion blur
+
+## Mitigations
+- Confidence thresholding
+- Human-in-the-loop fallback
+- Periodic retraining with licensed edge cases
+
+## Versioning
+- Model version: `v1.0.0`
+- Date: `YYYY-MM-DD`
+- Owner: `Company/Team`