v0.6.2: Gesture detection with HaGRID ONNX models (18 gestures)

PROJECT_PLAN.md

@@ -387,11 +387,18 @@ Based on deep analysis of Reachy Mini SDK, the following entities are exposed to
     - [x] AudioPlayer integrates aiosendspin library
     - [x] TTS audio sent to both local speaker and Sendspin server
 
+14. **Phase 22 - Gesture Detection** ✅ **Completed**
+    - [x] `gesture_detected` - Detected gesture name (Text Sensor)
+    - [x] `gesture_confidence` - Gesture detection confidence % (Sensor)
+    - [x] HaGRID ONNX models: hand_detector.onnx + crops_classifier.onnx
+    - [x] 18 supported gestures: call, dislike, fist, four, like, mute, ok, one, palm, peace, peace_inverted, rock, stop, stop_inverted, three, three2, two_up, two_up_inverted
+    - [x] Real-time state push to Home Assistant
+
 ---
 
-## 🎉 Phase 1-13 Entities Completed!
+## 🎉 Phase 1-13 + Phase 22 Entities Completed!
 
-**Total Completed: 43 entities**
+**Total Completed: 45 entities**
 - Phase 1: 4 entities (Basic status and volume)
 - Phase 2: 4 entities (Motor control)
 - Phase 3: 9 entities (Pose control)
@@ -405,6 +412,7 @@ Based on deep analysis of Reachy Mini SDK, the following entities are exposed to
 - Phase 11: 0 entities (LED control - Disabled)
 - Phase 12: 4 entities (Audio processing parameters)
 - Phase 13: 3 entities (Sendspin audio output)
+- Phase 22: 2 entities (Gesture detection)
 
 
 ---

changelog.json

@@ -1,4 +1,13 @@
 [
+  {
+    "version": "0.6.2",
+    "date": "2026-01-12",
+    "changes": [
+      "New: Gesture detection using HaGRID ONNX models (18 gesture classes)",
+      "New: gesture_detected and gesture_confidence entities in Home Assistant",
+      "Fix: Gesture state now properly pushed to Home Assistant in real-time"
+    ]
+  },
   {
     "version": "0.6.1",
     "date": "2026-01-12",

index.html

@@ -90,6 +90,11 @@
 				<h3>Sendspin Audio</h3>
 				<p>Multi-room audio support via Sendspin protocol. Sync playback with other speakers in your home.</p>
 			</div>
+			<div class="feature-card">
+				<span class="icon">🤚</span>
+				<h3>Gesture Detection</h3>
+				<p>HaGRID-based hand gesture recognition with 18 gestures like thumbs up, peace sign, and more.</p>
+			</div>
 		</div>
 	</section>
 

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "reachy_mini_ha_voice"
-version = "0.6.1"
+version = "0.6.2"
 description = "Home Assistant Voice Assistant for Reachy Mini"
 readme = "README.md"
 requires-python = ">=3.10"

reachy_mini_ha_voice/__init__.py

@@ -11,7 +11,7 @@ Key features:
 - Reachy Mini motion control integration
 """
 
-__version__ = "0.6.1"
+__version__ = "0.6.2"
 __author__ = "Desmond Dong"
 
 # Don't import main module here to avoid runpy warning

reachy_mini_ha_voice/camera_server.py

@@ -538,7 +538,7 @@ class MJPEGCameraServer:
                     self._gesture_confidence = confidence
                     if old_gesture != detected_gesture.value:
                         state_changed = True
-                        _LOGGER.info("Gesture detected: %s (%.1f%%)", 
+                        _LOGGER.debug("Gesture: %s (%.0f%%)", 
                                      detected_gesture.value, confidence * 100)
                 else:
                     if self._current_gesture != "none":
@@ -550,8 +550,8 @@ class MJPEGCameraServer:
             if state_changed and self._gesture_state_callback:
                 try:
                     self._gesture_state_callback()
-                except Exception as e:
-                    _LOGGER.debug("Gesture callback error: %s", e)
+                except Exception:
+                    pass  # Ignore callback errors
                     
         except Exception as e:
             _LOGGER.warning("Gesture detection error: %s", e)

reachy_mini_ha_voice/gesture_detector.py

@@ -166,13 +166,13 @@ class GestureDetector:
             if det is None:
                 return Gesture.NONE, 0.0
             x1, y1, x2, y2, det_c = det
-            logger.info("Hand detected: box=(%d,%d,%d,%d) conf=%.2f", x1, y1, x2, y2, det_c)
+            logger.debug("Hand: box=(%d,%d,%d,%d) conf=%.2f", x1, y1, x2, y2, det_c)
             crop = self._get_square_crop(frame, (x1, y1, x2, y2))
             if crop.size == 0:
                 return Gesture.NONE, 0.0
             gest, cls_c = self._classify(crop)
             if gest != Gesture.NONE:
-                logger.info("Gesture: %s (det=%.2f cls=%.2f)", gest.value, det_c, cls_c)
+                logger.debug("Gesture: %s (det=%.2f cls=%.2f)", gest.value, det_c, cls_c)
             return gest, det_c * cls_c
         except Exception as e:
             logger.warning("Gesture error: %s", e)