Previously SAM21 ran on the raw aligned depth and frequently leaked the
mask onto the table around the object: ~88% of segmented points (2136
of 2440 in a sample run) were table pixels filtered away by the post
partition_by_plane step. Open3DModel got a noisy cluster mostly off
the object.
New pipeline: with a calibrated work plane, depth pixels whose 3D
back-projection sits below the plane are zeroed BEFORE SAM21 sees the
depth PNG. SAM21 cannot back-project zero-depth pixels, so only true
above-plane points end up in the cloud regardless of mask outline.
Same sample run after the change: 513 SAM21 -> 513 accepted, 0 excluded.
The post-partition still runs (now a no-op on a calibrated plane), so
the find_model response shape is unchanged.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
points_via_sam21, _points_via_bbox, and _object_cloud_cv (bbox branch)
now call apply_below_plane_filter on the CV-frame cloud before the
<10-points guard. Responses carry plane_applied + n_points_before_plane.
Without a saved plane the filter is a no-op (plane_applied=false).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Owns WorkPlane (with new offset_mm field), fit_plane_svd, disk-backed
get/set, and the apply_below_plane_filter helper. WorkPlane.normal lives
in OpenCV RGB-camera frame and is oriented toward the camera. The filter
keeps points where signed_distance >= -offset_mm/1000.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Extract the SAM21 wiring out of grasp_routes.py into a shared module
perception/sam21_bridge.py (Sam21Fallback + points_via_sam21). Both
/api/perception/scan and /api/perception/object_points now try SAM21
first and fall back to the bbox ROI path on any Sam21Fallback. Response
now carries 'source' (sam21 | bbox_roi) and optional 'sam21_fallback_reason'.
Side-effect: 'points_camera' on both endpoints is now in OpenCV
RGB-camera frame (same as 'points_base' minus the T_base_camera
transform), regardless of which path produced the cloud. Previously the
bbox path stored points in Nuwa-native frame here. The web GUI plots
points_base when available, so the scatter views automatically pick up
the SAM21-segmented cloud.
Verified live: /scan → source=sam21 n=2414 sam21=111ms;
/object_points → source=sam21 n=2347 sam21=118ms;
/grasp/plan regression → source=sam21 n=2390 sam21=119ms;
/scan with SAM21 down → source=bbox_roi fallback='unreachable: ...'
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The 2× focal-length collision test wasn't actually colliding (adjacent IR
pixels rounded to different RGB columns). 4× makes them collapse, so the
np.minimum.at reduce is now exercised. Also drop unused H_ir, W_ir.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Forward-project IR depth pixels into the RGB pixel grid with a z-buffer.
Output uint16 mm at RGB resolution, ready to feed SAM21.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Pipeline completa per calibrazione hand-eye e grasp planning su robot
Unitree G1/H1 con camera depth Nuwa-HP60C. Esposta come FastAPI + GUI vanilla
JS, con assistente AI (OpenRouter + tool calling curato) per pilotare il
sistema in linguaggio naturale.
Architettura:
- FastAPI orchestrator (porta da .env, default 8890) che coordina 4
microservizi upstream: Nuwa-HP60C, marker-finder, YOLO, robot bridge.
- Dual-robot: services.g1 e services.h1 in config.yaml; switchabili a
runtime via PUT /api/home/robot.
- Calibrazione hand-eye eye-to-hand split per-side (right/left).
- Marker pose: marker-finder riceve gli intrinseci RGB della Nuwa al boot,
PnP server-side con quei dati. Nessun re-PnP locale richiesto.
- Depth-to-RGB: registrazione iterativa IR↔RGB per back-projection 3D
corretta (Nuwa /api/measure usa coords IR; il nostro /api/perception/
measure_rgb prende pixel RGB e ritorna 3D nel frame RGB OpenCV).
- Correzione affine sul depth bias (calibrabile via marker come
ground-truth) — riduce errore Z da ~6% a ~1.5% nei test.
- AI assistant: 16 tool curati (move_wrist_*, hand_action, get_marker[_3d],
detect_object, calibration_*, set_active_robot…), system prompt
dinamico che riflette robot attivo + marker config + addendum
customizzabile da GUI. Temperature di default 0.1 per evitare invenzioni.
GUI (7 tab):
- Home: stato comunicazione device + selettore robot attivo G1/H1
- Assistente AI: chat con OpenRouter, history persistita lato browser,
task riutilizzabili (prompt salvati), trace tool calls
- Impostazioni AI: modello da combo curata, temperature, max iterations,
prompt addendum, anteprima system prompt
- Calibrazione, Auto-Calibrazione, TrovaUtensile, Servo Visuale, Vista 3D,
Movimento EE — invariate dalle versioni precedenti
Setup:
- Variabili in presa-mano/.env (template in .env.example)
- OPENROUTER_API_KEY richiesta solo per il tab Assistente AI
- PRESA_MANO_PORT configurabile
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
