feat(right_arm): force, camera↔polso offset, cluster filter sul centroide

Tre miglioramenti al move-toward dopo il primo uso live:

1. checkbox "force" accanto a "Muovi": bypassa max_step_m quando
   l'utente sa che il delta richiesto e' grande ma corretto (es. primo
   avvicinamento all'oggetto).

2. camera_offset_wrist_mm: il modello assumeva camera e TCP coincidenti
   (sole sign flips). La camera Nuwa e' invece montata con traslazione
   costante rispetto al polso. La formula diventa:
       delta_wrist = sign * (centroid_cam - offset_cam) + t_wrist_cam
   Esposto in config (default 0,0,0), GUI (3 input "Camera rispetto a
   polso (mm)") e in /state. Il termine compensa il fatto che muovere
   il polso porta la camera, non il TCP, sul punto di presa.

3. filter_nearest_cluster: il centroide era media di tutti i punti
   SAM21 della mask. In presenza di sfondo lontano visibile attraverso
   buchi della depth del primo piano la media veniva trascinata a
   metri di distanza. Ora si tiene il cluster piu' vicino:
       z_near = p5(Z)
       keep   = punti con Z <= z_near + cluster_depth_window_m
   Default window 0.20 m, configurabile. La risposta di /detect/wrist
   espone n_sam21_points, n_centroid_points e z_near/z_cutoff per
   diagnosi.

Aggiornata anche la persistenza: /offset accetta camera_offset_wrist_mm.
GUI: i campi Offset/Sign/CamOff rispettano "user touched" (refresh non
li cancella mentre l'utente digita).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
Adriano Dal PastroandClaude Opus 4.7 committed 2026-05-21 12:18:24 +02:00
1 parent fcd921ff32
commit 1fa7d12ad7
8 files changed
+260 -20

No files matched your search

+32 -4
View File
@@ -15,6 +15,7 @@ from ..config import update_config
from ..geometry import quaternion_xyzw_to_rpy_deg, rpy_deg_to_quaternion_xyzw
from ..perception.detector import _normalize as _normalize_yolo
from ..perception.sam21_bridge import Sam21Fallback, segment_with_sam21
from ..right_arm.cluster import filter_nearest_cluster
from ..right_arm.move import DeltaOversize, apply_z_floor, compute_delta_wrist
from ..timing import Stopwatch
from . import _deps
@@ -40,6 +41,8 @@ def _state_payload(sw: Stopwatch) -> dict[str, Any]:
"custom_pose": cfg.get("custom_pose"),
"offset_camera_mm": cfg.get("move_offset_camera_mm"),
"axis_signs": cfg.get("axis_sign_cam_to_wrist"),
"camera_offset_wrist_mm": cfg.get("camera_offset_wrist_mm"),
"cluster_depth_window_m": cfg.get("cluster_depth_window_m"),
"max_step_m": cfg.get("max_step_m"),
"z_floor_m": cfg.get("z_floor_m"),
"timings_ms": sw.as_dict(),
@@ -158,9 +161,18 @@ def _detect_wrist_core(class_name: str | None, conf: float | None, sw: Stopwatch
seg = segment_with_sam21(nuwa, _deps.sam21(), bbox, sw)
except Sam21Fallback as e:
raise HTTPException(status_code=422, detail=f"SAM21 fallback: {e}")
pts = np.asarray(seg.points_cv_m, dtype=np.float64)
pts_all = np.asarray(seg.points_cv_m, dtype=np.float64)
if pts_all.shape[0] < 10:
raise HTTPException(status_code=422, detail=f"too few sam21 points: {pts_all.shape[0]}")
cfg = _deps.tab_right_arm_cfg()
window_m = float(cfg.get("cluster_depth_window_m", 0.20))
with sw.step("cluster_filter"):
pts, cluster_diag = filter_nearest_cluster(pts_all, window_m=window_m)
if pts.shape[0] < 10:
raise HTTPException(status_code=422, detail=f"too few sam21 points: {pts.shape[0]}")
raise HTTPException(
status_code=422,
detail=f"cluster troppo piccolo: {pts.shape[0]} punti dopo filtro "
f"(z_near={cluster_diag['z_near_m']:.3f}m window={window_m:.3f}m)")
with sw.step("centroid"):
centroid = pts.mean(axis=0)
distance = float(np.linalg.norm(centroid))
@@ -169,7 +181,11 @@ def _detect_wrist_core(class_name: str | None, conf: float | None, sw: Stopwatch
"bbox": list(bbox),
"centroid_camera_m": [float(centroid[0]), float(centroid[1]), float(centroid[2])],
"distance_m": distance,
"n_sam21_points": int(pts.shape[0]),
"n_sam21_points": int(pts_all.shape[0]),
"n_centroid_points": int(pts.shape[0]),
"cluster_window_m": window_m,
"cluster_z_near_m": cluster_diag["z_near_m"],
"cluster_z_cutoff_m": cluster_diag["z_cutoff_m"],
}
@@ -194,6 +210,7 @@ def move_toward(req: MoveTowardReq) -> dict[str, Any]:
cfg = _deps.tab_right_arm_cfg()
off_mm = cfg.get("move_offset_camera_mm", {"x": 0, "y": 0, "z": 0})
sign = cfg.get("axis_sign_cam_to_wrist", {"x": 1, "y": 1, "z": 1})
cam_off_mm = cfg.get("camera_offset_wrist_mm", {"x": 0, "y": 0, "z": 0})
max_step_m = float(cfg.get("max_step_m", 0.15))
z_floor = cfg.get("z_floor_m")
z_floor_m = float(z_floor) if z_floor is not None else None
@@ -201,10 +218,13 @@ def move_toward(req: MoveTowardReq) -> dict[str, Any]:
det = _detect_wrist_core(req.class_name, req.conf, sw)
centroid = np.asarray(det["centroid_camera_m"], dtype=np.float64)
offset_m = np.array([off_mm["x"], off_mm["y"], off_mm["z"]], dtype=np.float64) / 1000.0
cam_offset_wrist_m = np.array(
[cam_off_mm["x"], cam_off_mm["y"], cam_off_mm["z"]], dtype=np.float64) / 1000.0
signs = (int(sign["x"]), int(sign["y"]), int(sign["z"]))
try:
delta = compute_delta_wrist(centroid, offset_m, signs,
camera_offset_wrist_m=cam_offset_wrist_m,
max_step_m=max_step_m, force=req.force)
except DeltaOversize as e:
raise HTTPException(status_code=422, detail=str(e))
@@ -246,6 +266,7 @@ def move_toward(req: MoveTowardReq) -> dict[str, Any]:
"distance_m": det["distance_m"],
"offset_camera_m": offset_m.tolist(),
"axis_signs": list(signs),
"camera_offset_wrist_m": cam_offset_wrist_m.tolist(),
"delta_wrist_m": delta.tolist(),
"delta_wrist_mm": (delta * 1000.0).tolist(),
"z_clipped": bool(z_clipped),
@@ -269,6 +290,7 @@ class PoseUpdateReq(BaseModel):
class OffsetUpdateReq(BaseModel):
move_offset_camera_mm: Optional[dict[str, float]] = None
axis_sign_cam_to_wrist: Optional[dict[str, int]] = None
camera_offset_wrist_mm: Optional[dict[str, float]] = None
def _exec_named_pose(pose_key: str, dry_run: bool, sw: Stopwatch) -> dict[str, Any]:
@@ -332,9 +354,15 @@ def put_offset(req: OffsetUpdateReq) -> dict[str, Any]:
if v not in (-1, 1):
raise HTTPException(status_code=422, detail=f"axis_sign_cam_to_wrist['{k}'] must be +/-1")
patch["axis_sign_cam_to_wrist"] = {k: int(req.axis_sign_cam_to_wrist[k]) for k in ("x", "y", "z")}
if req.camera_offset_wrist_mm is not None:
for k in ("x", "y", "z"):
if k not in req.camera_offset_wrist_mm:
raise HTTPException(status_code=422, detail=f"camera_offset_wrist_mm missing '{k}'")
patch["camera_offset_wrist_mm"] = {k: float(req.camera_offset_wrist_mm[k]) for k in ("x", "y", "z")}
if not patch:
raise HTTPException(status_code=422, detail="nothing to update")
update_config({"tab_right_arm": patch})
cfg = _deps.tab_right_arm_cfg()
return {"move_offset_camera_mm": cfg.get("move_offset_camera_mm"),
"axis_sign_cam_to_wrist": cfg.get("axis_sign_cam_to_wrist")}
"axis_sign_cam_to_wrist": cfg.get("axis_sign_cam_to_wrist"),
"camera_offset_wrist_mm": cfg.get("camera_offset_wrist_mm")}