pollen-robotics · alozowski · Jul 8, 2026 · Jun 1, 2026 · Jun 1, 2026 · Jun 25, 2026
diff --git a/docs/source/API/openapi.json b/docs/source/API/openapi.json
@@ -1456,6 +1456,114 @@
         }
       }
     },
+    "/api/media/tracking/enable": {
+      "post": {
+        "summary": "Enable Tracking",
+        "description": "Enable daemon-side visual head tracking.",
+        "operationId": "enable_tracking_api_media_tracking_enable_post",
+        "requestBody": {
+          "content": {
+            "application/json": {
+              "schema": {
+                "anyOf": [
+                  {
+                    "$ref": "#/components/schemas/EnableTrackingRequest"
+                  },
+                  {
+                    "type": "null"
+                  }
+                ],
+                "title": "Body"
+              }
+            }
+          }
+        },
+        "responses": {
+          "200": {
+            "description": "Successful Response",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "additionalProperties": {
+                    "anyOf": [
+                      {
+                        "type": "string"
+                      },
+                      {
+                        "type": "boolean"
+                      }
+                    ]
+                  },
+                  "type": "object",
+                  "title": "Response Enable Tracking Api Media Tracking Enable Post"
+                }
+              }
+            }
+          },
+          "422": {
+            "description": "Validation Error",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/HTTPValidationError"
+                }
+              }
+            }
+          }
+        }
+      }
+    },
+    "/api/media/tracking/disable": {
+      "post": {
+        "summary": "Disable Tracking",
+        "description": "Disable daemon-side visual head tracking.",
+        "operationId": "disable_tracking_api_media_tracking_disable_post",
+        "responses": {
+          "200": {
+            "description": "Successful Response",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "additionalProperties": {
+                    "anyOf": [
+                      {
+                        "type": "string"
+                      },
+                      {
+                        "type": "boolean"
+                      }
+                    ]
+                  },
+                  "type": "object",
+                  "title": "Response Disable Tracking Api Media Tracking Disable Post"
+                }
+              }
+            }
+          }
+        }
+      }
+    },
+    "/api/media/tracking/face": {
+      "get": {
+        "summary": "Get Tracked Face",
+        "description": "Return the latest face observed by daemon-side head tracking.",
+        "operationId": "get_tracked_face_api_media_tracking_face_get",
+        "responses": {
+          "200": {
+            "description": "Successful Response",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "additionalProperties": true,
+                  "type": "object",
+                  "title": "Response Get Tracked Face Api Media Tracking Face Get"
+                }
+              }
+            }
+          }
+        }
+      }
+    },
     "/api/media/sounds/upload": {
       "post": {
         "summary": "Upload Sound",
@@ -2742,6 +2850,9 @@
               }
             ],
             "title": "Hardware Id"
+          },
+          "face_target": {
+            "$ref": "#/components/schemas/FaceTarget"
           }
         },
         "type": "object",
@@ -2776,6 +2887,76 @@
         "title": "DoAInfo",
         "description": "Direction of Arrival info from the microphone array."
       },
+      "EnableTrackingRequest": {
+        "properties": {
+          "weight": {
+            "type": "number",
+            "maximum": 1.0,
+            "minimum": 0.0,
+            "title": "Weight",
+            "default": 1.0
+          }
+        },
+        "type": "object",
+        "title": "EnableTrackingRequest",
+        "description": "Request body for enabling daemon-side head tracking."
+      },
+      "FaceTarget": {
+        "properties": {
+          "detected": {
+            "type": "boolean",
+            "title": "Detected",
+            "default": false
+          },
+          "x": {
+            "anyOf": [
+              {
+                "type": "number"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "title": "X"
+          },
+          "y": {
+            "anyOf": [
+              {
+                "type": "number"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "title": "Y"
+          },
+          "roll": {
+            "anyOf": [
+              {
+                "type": "number"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "title": "Roll"
+          },
+          "ts": {
+            "anyOf": [
+              {
+                "type": "number"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "title": "Ts"
+          }
+        },
+        "type": "object",
+        "title": "FaceTarget",
+        "description": "Latest daemon-side face target used for head tracking."
+      },
       "FullBodyTarget": {
         "properties": {
           "target_head_pose": {

diff --git a/docs/source/SDK/media-architecture.md b/docs/source/SDK/media-architecture.md
@@ -17,7 +17,7 @@ The daemon starts its media pipeline automatically unless the `--no-media` flag
 1. Opens the camera (platform-aware: v4l2, libcamera, DirectShow, AVFoundation, or UDP for simulation).
 2. Opens the audio device (platform-aware: PulseAudio, ALSA, WASAPI, CoreAudio).
 3. Feeds both into a WebRTC server (`webrtcsink`) for remote streaming.
-4. Exposes raw camera frames via a local IPC endpoint (`unixfdsink` on Linux/macOS, `win32ipcvideosink` on Windows).
+4. Exposes raw camera frames via a local IPC endpoint (`unixfdsink` on Linux/macOS, `win32ipcvideosink` on Windows). The IPC feed is served at a capped framerate (10 fps) below the capture rate: local vision clients convert every frame they are served, so the cap keeps on-device CPU low regardless of the camera mode.
 
 [![Reachy Mini Media Daemon](https://github.com/pollen-robotics/reachy_mini/raw/main/docs/assets/reachymini_media_daemon.png)]()
 

diff --git a/docs/source/SDK/python-sdk.md b/docs/source/SDK/python-sdk.md
@@ -44,6 +44,21 @@ with ReachyMini(media_backend="default") as mini:
 ```
 The returned frame is a numpy array with shape `(height, width, 3)` and data type `uint8`.
 
+### Head Tracking 👀
+
+The daemon can track the closest face and turn the head to follow it (aiming at the nose). Detection runs inside the daemon.
+
+```python
+from reachy_mini import ReachyMini
+
+with ReachyMini() as mini:
+    mini.start_head_tracking()
+    face = mini.get_tracked_face()  # detected, x, y in [-1, 1], roll
+    mini.stop_head_tracking()
+```
+
+`start_head_tracking(weight=...)` blends tracking with application motion: `1.0` lets tracking own the head, `0.0` pauses detection (freeing the head and CPU) without stopping the tracker, so applications can toggle it cheaply per turn. See the [Head Tracking example](../examples/head_tracking.md).
+
 ### IMU 🧭
 
 > ⚠️ The IMU is only available with the wireless version of Reachy Mini

diff --git a/docs/source/_toctree.yml b/docs/source/_toctree.yml
@@ -82,6 +82,8 @@
       title: Minimal demo
     - local: examples/look_at
       title: Look at
+    - local: examples/head_tracking
+      title: Head Tracking
     - local: examples/sequence
       title: Sequence
     - local: examples/goto_interpolation_playground

diff --git a/docs/source/examples/head_tracking.md b/docs/source/examples/head_tracking.md
@@ -0,0 +1,18 @@
+# Head Tracking
+
+This example enables daemon-side head tracking: Reachy Mini turns its head to follow the closest face, aiming at the nose. Detection runs inside the daemon (YuNet on ONNX Runtime), so the script only toggles tracking and polls the latest face target.
+
+Run with:
+```bash
+python head_tracking.py
+```
+
+`start_head_tracking(weight=...)` accepts a blend factor: `1.0` lets tracking own the head, `0.0` pauses detection (freeing the head and CPU) without stopping the tracker — useful to hand the head back to an application between conversation turns.
+
+<literalinclude>
+{"path": "../../../examples/head_tracking.py",
+"language": "python",
+"start-after": "START doc_example",
+"end-before": "END doc_example"
+}
+</literalinclude>
diff --git a/examples/head_tracking.py b/examples/head_tracking.py
@@ -0,0 +1,30 @@
+"""Enable daemon-side head tracking and print the tracked face target.
+
+Reachy Mini turns its head to follow the closest face. Detection runs inside
+the daemon, so this script only toggles tracking and reads the result.
+
+Note:
+    The daemon must be running before executing this script.
+
+"""
+
+# START doc_example
+
+from reachy_mini import ReachyMini
+
+with ReachyMini() as mini:
+    mini.start_head_tracking()
+    try:
+        while True:
+            # Waits for the next daemon status broadcast, so the loop runs at ~1 Hz.
+            face = mini.get_tracked_face()
+            if face.detected:
+                print(f"Face at x={face.x:+.2f}, y={face.y:+.2f}")
+            else:
+                print("No face detected")
+    except KeyboardInterrupt:
+        pass
+    finally:
+        mini.stop_head_tracking()
+
+# END doc_example
diff --git a/pyproject.toml b/pyproject.toml
@@ -8,38 +8,41 @@ version = "1.9.0rc1"
 authors = [{ name = "Pollen Robotics", email = "contact@pollen-robotics.com" }]
 description = ""
 readme = "README.md"
-requires-python = ">=3.10"
+requires-python = ">=3.11"
 dependencies = [
-    "numpy>=2.2.5",
-    "scipy>=1.15.3, <2.0.0",
-    "reachy_mini_motor_controller>=1.5.5",
-    "psutil",
-    "jinja2",
-    "uvicorn[standard]",
+    "aiohttp>=3.14.0",
+    "asgiref",
     "fastapi",
-    "pydantic>=2.12.5,<3.0.0",
-    "python-multipart>=0.0.27",
-    "starlette<1.0.0", # TODO: update TemplateResponse(request, name, context) before allowing Starlette 1.x
-    "pyserial",
     "huggingface-hub>=1.17.0,<2.0.0",
-    "reachy-mini-rust-kinematics>=1.0.3",
-    "asgiref",
-    "aiohttp>=3.14.0",
-    "log-throttling==0.0.3",
-    "pyusb>=1.2.1",
+    "jinja2",
     "libusb_package>=1.0.26.3",
+    "log-throttling==0.0.3",
+    "numpy>=2.2.5",
+    "onnxruntime==1.27.0",
     "pip>=26.1",
     "platformdirs",
-    "rich",
-    "questionary",
-    "websockets>=12,<16",
-    "toml",
-    "rustypot>=1.4.2",
+    "psutil",
+    "pulsectl>=24.1.0; sys_platform == 'linux'",
+    "pycaw>=20240210; sys_platform == 'win32'",
+    "pydantic>=2.12.5,<3.0.0",
+    "pyserial",
+    "python-multipart>=0.0.27",
+    "pyusb>=1.2.1",
     "pyyaml>=6.0",
+    "questionary",
     "requests>=2.33.0",
+    "rich",
+    "rustypot>=1.4.2",
+    "scipy>=1.15.3, <2.0.0",
+    "starlette<1.0.0", # TODO: update TemplateResponse(request, name, context) before allowing Starlette 1.x
+    "toml",
+    "uvicorn[standard]",
+    "websockets>=12,<16",
     "zeroconf>=0.149.7,<1",
-    "pycaw>=20240210; sys_platform == 'win32'",
-    "pulsectl>=24.1.0; sys_platform == 'linux'",
+
+    # Reachy Mini specifics
+    "reachy-mini-rust-kinematics>=1.0.3",
+    "reachy_mini_motor_controller>=1.5.5",
 
     "gstreamer-bundle==1.28.3; sys_platform != 'linux'",
     # OS-specific: use PyGObject for Linux
@@ -55,7 +58,6 @@ examples = [
     "gradio-client",
 ]
 mujoco = ["mujoco==3.3.0"]
-nn_kinematics = ["onnxruntime==1.22.1"]
 placo_kinematics = [
     "placo==0.9.14; sys_platform != 'win32'",
     "tornado>=6.5.5; sys_platform != 'win32'", # security floor for meshcat transitive dependency
@@ -76,7 +78,6 @@ opencv = ["opencv-python<=5.0", "cv2_enumerate_cameras>=1.2.1"]
 all = [
     "reachy_mini[examples]",
     "reachy_mini[mujoco]",
-    "reachy_mini[nn_kinematics]",
     "reachy_mini[placo_kinematics]",
     "reachy_mini[rerun]",
     "reachy_mini[wireless-version]",