Merge branch 'main' into feat/face-recognition

ref: N25B-397

pyproject.toml

@@ -7,6 +7,7 @@ requires-python = ">=3.13"
 dependencies = [
   "agentspeak>=0.2.2",
   "colorlog>=6.10.1",
+  "deepface>=0.0.96",
   "fastapi[all]>=0.115.6",
   "mlx-whisper>=0.4.3 ; sys_platform == 'darwin'",
   "numpy>=2.3.3",
@@ -21,6 +22,7 @@ dependencies = [
   "silero-vad>=6.0.0",
   "sphinx>=7.3.7",
   "sphinx-rtd-theme>=3.0.2",
+  "tf-keras>=2.20.1",
   "torch>=2.8.0",
   "uvicorn>=0.37.0",
 ]

src/control_backend/agents/bdi/agentspeak_generator.py

@@ -29,6 +29,8 @@ from control_backend.schemas.program import (
     BaseGoal,
     BasicNorm,
     ConditionalNorm,
+    EmotionBelief,
+    FaceBelief,
     GestureAction,
     Goal,
     InferredBelief,
@@ -682,6 +684,14 @@ class AgentSpeakGenerator:
         """
         return AstLiteral(self.slugify(sb))
 
+    @_astify.register
+    def _(self, eb: EmotionBelief) -> AstExpression:
+        return AstLiteral("emotion_detected", [AstAtom(eb.emotion)])
+
+    @_astify.register
+    def _(self, eb: FaceBelief) -> AstExpression:
+        return AstLiteral("face_present")
+
     @_astify.register
     def _(self, ib: InferredBelief) -> AstExpression:
         """

src/control_backend/agents/communication/ri_communication_agent.py

@@ -9,17 +9,17 @@ import json
 
 import zmq
 import zmq.asyncio as azmq
-from pydantic import ValidationError
 from zmq.asyncio import Context
 
 from control_backend.agents import BaseAgent
 from control_backend.agents.actuation.robot_gesture_agent import RobotGestureAgent
+from control_backend.agents.perception.visual_emotion_recognition_agent.visual_emotion_recognition_agent import (  # noqa
+    VisualEmotionRecognitionAgent,
+)
 from control_backend.core.config import settings
-from control_backend.schemas.internal_message import InternalMessage
-from control_backend.schemas.ri_message import PauseCommand
 
 from ..actuation.robot_speech_agent import RobotSpeechAgent
-from ..perception import VADAgent
+from ..perception import FacePerceptionAgent, VADAgent
 
 
 class RICommunicationAgent(BaseAgent):
@@ -58,6 +58,7 @@ class RICommunicationAgent(BaseAgent):
         self.connected = False
         self.gesture_agent: RobotGestureAgent | None = None
         self.speech_agent: RobotSpeechAgent | None = None
+        self.visual_emotion_recognition_agent: VisualEmotionRecognitionAgent | None = None
 
     async def setup(self):
         """
@@ -180,7 +181,7 @@ class RICommunicationAgent(BaseAgent):
             bind = port_data["bind"]
 
             if not bind:
-                addr = f"tcp://{settings.ri_host}:{port}"
+                addr = f"tcp://localhost:{port}"
             else:
                 addr = f"tcp://*:{port}"
 
@@ -215,6 +216,21 @@ class RICommunicationAgent(BaseAgent):
                 case "audio":
                     vad_agent = VADAgent(audio_in_address=addr, audio_in_bind=bind)
                     await vad_agent.start()
+                case "video":
+                    visual_emotion_agent = VisualEmotionRecognitionAgent(
+                        settings.agent_settings.visual_emotion_recognition_name,
+                        socket_address=addr,
+                        bind=bind,
+                    )
+                    self.visual_emotion_recognition_agent = visual_emotion_agent
+                    await visual_emotion_agent.start()
+                case "face":
+                    face_agent = FacePerceptionAgent(
+                        settings.agent_settings.face_agent_name,
+                        zmq_address=addr,
+                        zmq_bind=bind,
+                    )
+                    await face_agent.start()
                 case _:
                     self.logger.warning("Unhandled negotiation id: %s", id)
 
@@ -320,17 +336,12 @@ class RICommunicationAgent(BaseAgent):
         if self.speech_agent is not None:
             await self.speech_agent.stop()
 
+        if self.visual_emotion_recognition_agent is not None:
+            await self.visual_emotion_recognition_agent.stop()
+
         if self.pub_socket is not None:
             self.pub_socket.close()
 
         self.logger.debug("Restarting communication negotiation.")
         if await self._negotiate_connection(max_retries=2):
             self.connected = True
-
-    async def handle_message(self, msg: InternalMessage):
-        try:
-            pause_command = PauseCommand.model_validate_json(msg.body)
-            await self._req_socket.send_json(pause_command.model_dump())
-            self.logger.debug(await self._req_socket.recv_json())
-        except ValidationError:
-            self.logger.warning("Incorrect message format for PauseCommand.")

src/control_backend/agents/perception/__init__.py

@@ -7,6 +7,7 @@ Agents responsible for processing sensory input, such as audio transcription and
 detection.
 """
 
+from .face_rec_agent import FacePerceptionAgent as FacePerceptionAgent
 from .transcription_agent.transcription_agent import (
     TranscriptionAgent as TranscriptionAgent,
 )

src/control_backend/agents/perception/face_rec_agent.py

@@ -0,0 +1,144 @@
+"""
+This program has been developed by students from the bachelor Computer Science at Utrecht
+University within the Software Project course.
+© Copyright Utrecht University (Department of Information and Computing Sciences)
+"""
+
+import asyncio
+
+import zmq
+import zmq.asyncio as azmq
+
+from control_backend.agents import BaseAgent
+from control_backend.core.agent_system import InternalMessage
+from control_backend.core.config import settings
+from control_backend.schemas.belief_message import Belief, BeliefMessage
+
+
+class FacePerceptionAgent(BaseAgent):
+    """
+    Receives face presence updates from the RICommunicationAgent
+    via the internal PUB/SUB bus.
+    """
+
+    def __init__(self, name: str, zmq_address: str, zmq_bind: bool):
+        """
+        :param name: The name of the agent.
+        :param zmq_address: The ZMQ address to subscribe to, an endpoint which sends face presence
+            updates.
+        :param zmq_bind: Whether to connect to the ZMQ endpoint, or to bind.
+        """
+        super().__init__(name)
+        self._zmq_address = zmq_address
+        self._zmq_bind = zmq_bind
+        self._socket: azmq.Socket | None = None
+
+        self._last_face_state: bool | None = None
+
+        # Pause functionality
+        # NOTE: flag is set when running, cleared when paused
+        self._paused = asyncio.Event()
+        self._paused.set()
+
+    async def setup(self):
+        self.logger.info("Starting FacePerceptionAgent")
+
+        if self._socket is None:
+            self._connect_socket()
+
+        self.add_behavior(self._poll_loop())
+        self.logger.info("Finished setting up %s", self.name)
+
+    def _connect_socket(self):
+        if self._socket is not None:
+            self.logger.warning("ZMQ socket already initialized. Did you call setup() twice?")
+            return
+
+        self._socket = azmq.Context.instance().socket(zmq.SUB)
+        self._socket.setsockopt_string(zmq.SUBSCRIBE, "")
+        if self._zmq_bind:
+            self._socket.bind(self._zmq_address)
+        else:
+            self._socket.connect(self._zmq_address)
+
+    async def _poll_loop(self):
+        if self._socket is None:
+            self.logger.warning("Connection not initialized before poll loop. Call setup() first.")
+            return
+
+        while self._running:
+            try:
+                await self._paused.wait()
+                response = await asyncio.wait_for(
+                    self._socket.recv_json(), timeout=settings.behaviour_settings.sleep_s
+                )
+
+                face_present = response.get("face_detected", False)
+
+                if self._last_face_state is None:
+                    self._last_face_state = face_present
+                    continue
+
+                if face_present != self._last_face_state:
+                    self._last_face_state = face_present
+                    self.logger.debug("Face detected" if face_present else "Face lost")
+                    await self._update_face_belief(face_present)
+            except TimeoutError:
+                pass
+            except Exception as e:
+                self.logger.error("Face polling failed", exc_info=e)
+
+    async def _post_face_belief(self, present: bool):
+        """
+        Send a face_present belief update to the BDI Core Agent.
+        """
+        if present:
+            belief_msg = BeliefMessage(create=[{"name": "face_present", "arguments": []}])
+        else:
+            belief_msg = BeliefMessage(delete=[{"name": "face_present", "arguments": []}])
+
+        msg = InternalMessage(
+            to=settings.agent_settings.bdi_core_name,
+            sender=self.name,
+            thread="beliefs",
+            body=belief_msg.model_dump_json(),
+        )
+
+        await self.send(msg)
+
+    async def _update_face_belief(self, present: bool):
+        """
+        Add or remove the `face_present` belief in the BDI Core Agent.
+        """
+        if present:
+            payload = BeliefMessage(create=[Belief(name="face_present").model_dump()])
+        else:
+            payload = BeliefMessage(delete=[Belief(name="face_present").model_dump()])
+
+        message = InternalMessage(
+            to=settings.agent_settings.bdi_core_name,
+            sender=self.name,
+            thread="beliefs",
+            body=payload.model_dump_json(),
+        )
+
+        await self.send(message)
+
+    async def handle_message(self, msg: InternalMessage):
+        """
+        Handle incoming pause/resume commands from User Interrupt Agent.
+        """
+        sender = msg.sender
+
+        if sender == settings.agent_settings.user_interrupt_name:
+            if msg.body == "PAUSE":
+                self.logger.info("Pausing Face Perception processing.")
+                self._paused.clear()
+                self._last_face_state = None
+            elif msg.body == "RESUME":
+                self.logger.info("Resuming Face Perception processing.")
+                self._paused.set()
+            else:
+                self.logger.warning("Unknown command from User Interrupt Agent: %s", msg.body)
+        else:
+            self.logger.debug("Ignoring message from unknown sender: %s", sender)

src/control_backend/agents/perception/visual_emotion_recognition_agent/visual_emotion_recognition_agent.py

@@ -0,0 +1,207 @@
+import asyncio
+import json
+import time
+from collections import Counter, defaultdict
+
+import cv2
+import numpy as np
+import zmq
+import zmq.asyncio as azmq
+from pydantic_core import ValidationError
+
+from control_backend.agents import BaseAgent
+from control_backend.agents.perception.visual_emotion_recognition_agent.visual_emotion_recognizer import (  # noqa
+    DeepFaceEmotionRecognizer,
+)
+from control_backend.core.agent_system import InternalMessage
+from control_backend.core.config import settings
+from control_backend.schemas.belief_message import Belief
+
+
+class VisualEmotionRecognitionAgent(BaseAgent):
+    def __init__(
+        self,
+        name: str,
+        socket_address: str,
+        bind: bool = False,
+        timeout_ms: int = 1000,
+        window_duration: int = settings.behaviour_settings.visual_emotion_recognition_window_duration_s,  # noqa
+        min_frames_required: int = settings.behaviour_settings.visual_emotion_recognition_min_frames_per_face,  # noqa
+    ):
+        """
+        Initialize the Visual Emotion Recognition Agent.
+
+        :param name: Name of the agent
+        :param socket_address: Address of the socket to connect or bind to
+        :param bind: Whether to bind to the socket address (True) or connect (False)
+        :param timeout_ms: Timeout for socket receive operations in milliseconds
+        :param window_duration: Duration in seconds over which to aggregate emotions
+        :param min_frames_required: Minimum number of frames per face required to consider a face
+        valid
+        """
+        super().__init__(name)
+        self.socket_address = socket_address
+        self.socket_bind = bind
+        self.timeout_ms = timeout_ms
+        self.window_duration = window_duration
+        self.min_frames_required = min_frames_required
+
+        # Pause functionality
+        # NOTE: flag is set when running, cleared when paused
+        self._paused = asyncio.Event()
+        self._paused.set()
+
+    async def setup(self):
+        """
+        Initialize the agent resources.
+        1. Initializes the :class:`VisualEmotionRecognizer`.
+        2. Connects to the video input ZMQ socket.
+        3. Starts the background emotion recognition loop.
+        """
+        self.logger.info("Setting up %s.", self.name)
+
+        self.emotion_recognizer = DeepFaceEmotionRecognizer()
+
+        self.video_in_socket = azmq.Context.instance().socket(zmq.SUB)
+
+        if self.socket_bind:
+            self.video_in_socket.bind(self.socket_address)
+        else:
+            self.video_in_socket.connect(self.socket_address)
+
+        self.video_in_socket.setsockopt_string(zmq.SUBSCRIBE, "")
+        self.video_in_socket.setsockopt(zmq.RCVTIMEO, self.timeout_ms)
+        self.video_in_socket.setsockopt(zmq.CONFLATE, 1)
+
+        self.add_behavior(self.emotion_update_loop())
+
+        self.logger.info("Finished setting up %s", self.name)
+
+    async def emotion_update_loop(self):
+        """
+        Background loop to receive video frames, recognize emotions, and update beliefs.
+        1. Receives video frames from the ZMQ socket.
+        2. Uses the :class:`VisualEmotionRecognizer` to detect emotions.
+        3. Aggregates emotions over a time window.
+        4. Sends updates to the BDI Core Agent about detected emotions.
+        """
+        # Next time to process the window and update emotions
+        next_window_time = time.time() + self.window_duration
+
+        # Tracks counts of detected emotions per face index
+        face_stats = defaultdict(Counter)
+
+        prev_dominant_emotions = set()
+
+        while self._running:
+            try:
+                await self._paused.wait()
+
+                frame_bytes = await self.video_in_socket.recv()
+
+                # Convert bytes to a numpy buffer
+                nparr = np.frombuffer(frame_bytes, np.uint8)
+
+                # Decode image into the generic Numpy Array DeepFace expects
+                frame_image = cv2.imdecode(nparr, cv2.IMREAD_COLOR)
+
+                if frame_image is None:
+                    # Could not decode image, skip this frame
+                    self.logger.warning("Received invalid video frame, skipping.")
+                    continue
+
+                # Get the dominant emotion from each face
+                current_emotions = self.emotion_recognizer.sorted_dominant_emotions(frame_image)
+                # Update emotion counts for each detected face
+                for i, emotion in enumerate(current_emotions):
+                    face_stats[i][emotion] += 1
+
+                # If window duration has passed, process the collected stats
+                if time.time() >= next_window_time:
+                    window_dominant_emotions = set()
+                    # Determine dominant emotion for each face in the window
+                    for _, counter in face_stats.items():
+                        total_detections = sum(counter.values())
+
+                        if total_detections >= self.min_frames_required:
+                            dominant_emotion = counter.most_common(1)[0][0]
+                            window_dominant_emotions.add(dominant_emotion)
+
+                    await self.update_emotions(prev_dominant_emotions, window_dominant_emotions)
+                    prev_dominant_emotions = window_dominant_emotions
+                    face_stats.clear()
+                    next_window_time = time.time() + self.window_duration
+
+            except zmq.Again:
+                self.logger.warning("No video frame received within timeout.")
+
+    async def update_emotions(self, prev_emotions: set[str], emotions: set[str]):
+        """
+        Compare emotions from previous window and current emotions,
+        send updates to BDI Core Agent.
+        """
+        emotions_to_remove = prev_emotions - emotions
+        emotions_to_add = emotions - prev_emotions
+
+        if not emotions_to_add and not emotions_to_remove:
+            return
+
+        emotion_beliefs_remove = []
+        for emotion in emotions_to_remove:
+            self.logger.info(f"Emotion '{emotion}' has disappeared.")
+            try:
+                emotion_beliefs_remove.append(
+                    Belief(name="emotion_detected", arguments=[emotion], remove=True)
+                )
+            except ValidationError:
+                self.logger.warning("Invalid belief for emotion removal: %s", emotion)
+
+        emotion_beliefs_add = []
+        for emotion in emotions_to_add:
+            self.logger.info(f"New emotion detected: '{emotion}'")
+            try:
+                emotion_beliefs_add.append(Belief(name="emotion_detected", arguments=[emotion]))
+            except ValidationError:
+                self.logger.warning("Invalid belief for new emotion: %s", emotion)
+
+        beliefs_list_add = [b.model_dump() for b in emotion_beliefs_add]
+        beliefs_list_remove = [b.model_dump() for b in emotion_beliefs_remove]
+        payload = {"create": beliefs_list_add, "delete": beliefs_list_remove}
+
+        message = InternalMessage(
+            to=settings.agent_settings.bdi_core_name,
+            sender=self.name,
+            body=json.dumps(payload),
+            thread="beliefs",
+        )
+        await self.send(message)
+
+    async def handle_message(self, msg: InternalMessage):
+        """
+        Handle incoming messages.
+
+        Expects messages to pause or resume the Visual Emotion Recognition
+        processing from User Interrupt Agent.
+
+        :param msg: The received internal message.
+        """
+        sender = msg.sender
+
+        if sender == settings.agent_settings.user_interrupt_name:
+            if msg.body == "PAUSE":
+                self.logger.info("Pausing Visual Emotion Recognition processing.")
+                self._paused.clear()
+            elif msg.body == "RESUME":
+                self.logger.info("Resuming Visual Emotion Recognition processing.")
+                self._paused.set()
+            else:
+                self.logger.warning(f"Unknown command from User Interrupt Agent: {msg.body}")
+        else:
+            self.logger.debug(f"Ignoring message from unknown sender: {sender}")
+
+    async def stop(self):
+        """
+        Clean up resources used by the agent.
+        """
+        self.video_in_socket.close()
+        await super().stop()

src/control_backend/agents/perception/visual_emotion_recognition_agent/visual_emotion_recognizer.py

@@ -0,0 +1,54 @@
+import abc
+
+import numpy as np
+from deepface import DeepFace
+
+
+class VisualEmotionRecognizer(abc.ABC):
+    @abc.abstractmethod
+    def load_model(self):
+        """Load the visual emotion recognition model into memory."""
+        pass
+
+    @abc.abstractmethod
+    def sorted_dominant_emotions(self, image) -> list[str]:
+        """
+        Recognize dominant emotions from faces in the given image.
+        Emotions can be one of ['angry', 'disgust', 'fear', 'happy', 'sad', 'surprise', 'neutral'].
+        To minimize false positives, consider filtering faces with low confidence.
+
+        :param image: The input image for emotion recognition.
+        :return: List of dominant emotion detected for each face in the image,
+        sorted per face.
+        """
+        pass
+
+
+class DeepFaceEmotionRecognizer(VisualEmotionRecognizer):
+    """
+    DeepFace-based implementation of VisualEmotionRecognizer.
+    DeepFape has proven to be quite a pessimistic model, so expect sad, fear and neutral
+    emotions to be over-represented.
+    """
+
+    def __init__(self):
+        self.load_model()
+
+    def load_model(self):
+        print("Loading Deepface Emotion Model...")
+        dummy_img = np.zeros((224, 224, 3), dtype=np.uint8)
+        # analyze does not take a model as an argument, calling it once on a dummy image to load
+        # the model
+        DeepFace.analyze(dummy_img, actions=["emotion"], enforce_detection=False)
+        print("Deepface Emotion Model loaded.")
+
+    def sorted_dominant_emotions(self, image) -> list[str]:
+        analysis = DeepFace.analyze(image, actions=["emotion"], enforce_detection=False)
+
+        # Sort faces by x coordinate to maintain left-to-right order
+        analysis.sort(key=lambda face: face["region"]["x"])
+
+        analysis = [face for face in analysis if face["face_confidence"] >= 0.90]
+
+        dominant_emotions = [face["dominant_emotion"] for face in analysis]
+        return dominant_emotions

src/control_backend/agents/user_interrupt/user_interrupt_agent.py

@@ -18,7 +18,6 @@ from control_backend.schemas.belief_message import Belief, BeliefMessage
 from control_backend.schemas.program import ConditionalNorm, Goal, Program
 from control_backend.schemas.ri_message import (
     GestureCommand,
-    PauseCommand,
     RIEndpoint,
     SpeechCommand,
 )
@@ -398,34 +397,35 @@ class UserInterruptAgent(BaseAgent):
         self.logger.debug("Sending experiment control '%s' to BDI Core.", thread)
         await self.send(out_msg)
 
-    async def _send_pause_command(self, pause):
+    async def _send_pause_command(self, pause: str):
         """
-        Send a pause command to the Robot Interface via the RI Communication Agent.
+        Send a pause command to the other internal agents; for now just VAD and VED agent.
-        Send a pause command to the other internal agents; for now just VAD agent.
         """
-        cmd = PauseCommand(data=pause)
-        message = InternalMessage(
-            to=settings.agent_settings.ri_communication_name,
-            sender=self.name,
-            body=cmd.model_dump_json(),
-        )
-        await self.send(message)
-
         if pause == "true":
-            # Send pause to VAD agent
+            # Send pause to VAD and VED agent
             vad_message = InternalMessage(
-                to=settings.agent_settings.vad_name,
+                to=[
+                    settings.agent_settings.vad_name,
+                    settings.agent_settings.visual_emotion_recognition_name,
+                    settings.agent_settings.face_agent_name,
+                ],
                 sender=self.name,
                 body="PAUSE",
             )
             await self.send(vad_message)
-            self.logger.info("Sent pause command to VAD Agent and RI Communication Agent.")
+            # Voice Activity Detection and Visual Emotion Recognition agents
+            self.logger.info("Sent pause command to perception agents.")
         else:
-            # Send resume to VAD agent
+            # Send resume to VAD and VED agents
             vad_message = InternalMessage(
-                to=settings.agent_settings.vad_name,
+                to=[
+                    settings.agent_settings.vad_name,
+                    settings.agent_settings.visual_emotion_recognition_name,
+                    settings.agent_settings.face_agent_name,
+                ],
                 sender=self.name,
                 body="RESUME",
             )
             await self.send(vad_message)
-            self.logger.info("Sent resume command to VAD Agent and RI Communication Agent.")
+            # Voice Activity Detection and Visual Emotion Recognition agents
+            self.logger.info("Sent resume command to perception agents.")

src/control_backend/api/v1/router.py

@@ -12,7 +12,7 @@ api_router = APIRouter()
 
 api_router.include_router(message.router, tags=["Messages"])
 
-api_router.include_router(robot.router, prefix="/robot", tags=["Pings", "Commands"])
+api_router.include_router(robot.router, prefix="/robot", tags=["Pings", "Commands", "Face"])
 
 api_router.include_router(logs.router, tags=["Logs"])
 

src/control_backend/core/config.py

@@ -54,6 +54,7 @@ class AgentSettings(BaseModel):
     # agent names
     bdi_core_name: str = "bdi_core_agent"
     bdi_program_manager_name: str = "bdi_program_manager_agent"
+    visual_emotion_recognition_name: str = "visual_emotion_recognition_agent"
     text_belief_extractor_name: str = "text_belief_extractor_agent"
     vad_name: str = "vad_agent"
     llm_name: str = "llm_agent"
@@ -63,6 +64,7 @@ class AgentSettings(BaseModel):
     robot_speech_name: str = "robot_speech_agent"
     robot_gesture_name: str = "robot_gesture_agent"
     user_interrupt_name: str = "user_interrupt_agent"
+    face_agent_name: str = "face_detection_agent"
 
 
 class BehaviourSettings(BaseModel):
@@ -83,6 +85,10 @@ class BehaviourSettings(BaseModel):
     :ivar conversation_history_length_limit: The maximum amount of messages to extract beliefs from.
     :ivar trigger_time_to_wait: Amount of milliseconds to wait before informing the UI about trigger
      completion.
+    :ivar visual_emotion_recognition_window_duration_s: Duration in seconds over which to aggregate
+    emotions and update emotion beliefs.
+    :ivar visual_emotion_recognition_min_frames_per_face: Minimum number of frames per face required
+    to consider a face valid.
     """
 
     # ATTENTION: When adding/removing settings, make sure to update the .env.example file
@@ -110,6 +116,10 @@ class BehaviourSettings(BaseModel):
     trigger_time_to_wait: int = 2000
     agentspeak_file: str = "src/control_backend/agents/bdi/agentspeak.asl"
 
+    # Visual Emotion Recognition settings
+    visual_emotion_recognition_window_duration_s: int = 5
+    visual_emotion_recognition_min_frames_per_face: int = 3
+
 
 class LLMSettings(BaseModel):
     """

src/control_backend/schemas/program.py

@@ -41,8 +41,8 @@ class LogicalOperator(Enum):
     OR = "OR"
 
 
-type Belief = KeywordBelief | SemanticBelief | InferredBelief
+type Belief = KeywordBelief | SemanticBelief | InferredBelief | EmotionBelief | FaceBelief
-type BasicBelief = KeywordBelief | SemanticBelief
+type BasicBelief = KeywordBelief | SemanticBelief | EmotionBelief | FaceBelief
 
 
 class KeywordBelief(ProgramElement):
@@ -106,6 +106,27 @@ class InferredBelief(ProgramElement):
     right: Belief
 
 
+class EmotionBelief(ProgramElement):
+    """
+    Represents a belief that is set when a certain emotion is detected.
+
+    :ivar emotion: The emotion on which this belief gets set.
+    """
+
+    name: str = ""
+    emotion: str
+
+
+class FaceBelief(ProgramElement):
+    """
+    Represents the belief that at least one face is currently detected.
+    This belief is maintained by a perception agent (not inferred).
+    """
+
+    face_present: bool
+    name: str = ""
+
+
 class Norm(ProgramElement):
     """
     Base class for behavioral norms that guide the robot's interactions.
@@ -315,3 +336,9 @@ class Program(BaseModel):
     """
 
     phases: list[Phase]
+
+
+if __name__ == "__main__":
+    input = input("Enter program JSON: ")
+    program = Program.model_validate_json(input)
+    print(program)

test/unit/agents/communication/test_ri_communication_agent.py

@@ -10,8 +10,6 @@ from unittest.mock import ANY, AsyncMock, MagicMock, patch
 import pytest
 
 from control_backend.agents.communication.ri_communication_agent import RICommunicationAgent
-from control_backend.core.agent_system import InternalMessage
-from control_backend.schemas.ri_message import PauseCommand, RIEndpoint
 
 
 def speech_agent_path():
@@ -402,38 +400,3 @@ async def test_negotiate_req_socket_none_causes_retry(zmq_context):
     result = await agent._negotiate_connection(max_retries=1)
 
     assert result is False
-
-
-@pytest.mark.asyncio
-async def test_handle_message_pause_command(zmq_context):
-    """Test handle_message with a valid PauseCommand."""
-    agent = RICommunicationAgent("ri_comm")
-    agent._req_socket = AsyncMock()
-    agent.logger = MagicMock()
-
-    agent._req_socket.recv_json.return_value = {"status": "ok"}
-
-    pause_cmd = PauseCommand(data=True)
-    msg = InternalMessage(to="ri_comm", sender="user_int", body=pause_cmd.model_dump_json())
-
-    await agent.handle_message(msg)
-
-    agent._req_socket.send_json.assert_awaited_once()
-    args = agent._req_socket.send_json.await_args[0][0]
-    assert args["endpoint"] == RIEndpoint.PAUSE.value
-    assert args["data"] is True
-
-
-@pytest.mark.asyncio
-async def test_handle_message_invalid_pause_command(zmq_context):
-    """Test handle_message with invalid JSON."""
-    agent = RICommunicationAgent("ri_comm")
-    agent._req_socket = AsyncMock()
-    agent.logger = MagicMock()
-
-    msg = InternalMessage(to="ri_comm", sender="user_int", body="invalid json")
-
-    await agent.handle_message(msg)
-
-    agent.logger.warning.assert_called_with("Incorrect message format for PauseCommand.")
-    agent._req_socket.send_json.assert_not_called()

test/unit/agents/perception/test_face_detection_agent.py

@@ -0,0 +1,152 @@
+from unittest.mock import AsyncMock, MagicMock
+
+import pytest
+import zmq
+
+import control_backend.agents.perception.face_rec_agent as face_module
+from control_backend.agents.perception.face_rec_agent import FacePerceptionAgent
+from control_backend.core.agent_system import InternalMessage
+from control_backend.schemas.belief_message import BeliefMessage
+
+
+@pytest.fixture
+def agent():
+    """Return a FacePerceptionAgent instance for testing."""
+    return FacePerceptionAgent(
+        name="face_agent",
+        zmq_address="inproc://test",
+        zmq_bind=False,
+    )
+
+
+@pytest.fixture
+def socket():
+    """Return a mocked ZMQ socket."""
+    sock = AsyncMock()
+    sock.setsockopt_string = MagicMock()
+    sock.connect = MagicMock()
+    sock.bind = MagicMock()
+    return sock
+
+
+def test_connect_socket_connect(agent, socket, monkeypatch):
+    """Test that _connect_socket properly connects when zmq_bind=False."""
+    ctx = MagicMock()
+    ctx.socket.return_value = socket
+    monkeypatch.setattr(face_module.azmq, "Context", MagicMock(instance=lambda: ctx))
+
+    agent._connect_socket()
+    socket.setsockopt_string.assert_called_once_with(zmq.SUBSCRIBE, "")
+    socket.connect.assert_called_once_with(agent._zmq_address)
+    socket.bind.assert_not_called()
+
+
+def test_connect_socket_bind(agent, socket, monkeypatch):
+    """Test that _connect_socket properly binds when zmq_bind=True."""
+    agent._zmq_bind = True
+    ctx = MagicMock()
+    ctx.socket.return_value = socket
+    monkeypatch.setattr(face_module.azmq, "Context", MagicMock(instance=lambda: ctx))
+
+    agent._connect_socket()
+    socket.bind.assert_called_once_with(agent._zmq_address)
+    socket.connect.assert_not_called()
+
+
+def test_connect_socket_twice_is_noop(agent, socket):
+    """Test that calling _connect_socket twice does not overwrite an existing socket."""
+    agent._socket = socket
+    agent._connect_socket()
+    assert agent._socket is socket
+
+
+@pytest.mark.asyncio
+async def test_update_face_belief_present(agent):
+    """Test that _update_face_belief(True) creates the 'face_present' belief."""
+    agent.send = AsyncMock()
+    await agent._update_face_belief(True)
+    msg = agent.send.await_args.args[0]
+    payload = BeliefMessage.model_validate_json(msg.body)
+    assert payload.create[0].name == "face_present"
+
+
+@pytest.mark.asyncio
+async def test_update_face_belief_absent(agent):
+    """Test that _update_face_belief(False) deletes the 'face_present' belief."""
+    agent.send = AsyncMock()
+    await agent._update_face_belief(False)
+    msg = agent.send.await_args.args[0]
+    payload = BeliefMessage.model_validate_json(msg.body)
+    assert payload.delete[0].name == "face_present"
+
+
+@pytest.mark.asyncio
+async def test_post_face_belief_present(agent):
+    """Test that _post_face_belief(True) sends a belief creation message."""
+    agent.send = AsyncMock()
+    await agent._post_face_belief(True)
+    msg = agent.send.await_args.args[0]
+    assert '"create"' in msg.body and '"face_present"' in msg.body
+
+
+@pytest.mark.asyncio
+async def test_post_face_belief_absent(agent):
+    """Test that _post_face_belief(False) sends a belief deletion message."""
+    agent.send = AsyncMock()
+    await agent._post_face_belief(False)
+    msg = agent.send.await_args.args[0]
+    assert '"delete"' in msg.body and '"face_present"' in msg.body
+
+
+@pytest.mark.asyncio
+async def test_handle_pause(agent):
+    """Test that a 'PAUSE' message clears _paused and resets _last_face_state."""
+    agent._paused.set()
+    agent._last_face_state = True
+    msg = InternalMessage(
+        to=agent.name,
+        sender=face_module.settings.agent_settings.user_interrupt_name,
+        thread="cmd",
+        body="PAUSE",
+    )
+    await agent.handle_message(msg)
+    assert not agent._paused.is_set()
+    assert agent._last_face_state is None
+
+
+@pytest.mark.asyncio
+async def test_handle_resume(agent):
+    """Test that a 'RESUME' message sets _paused."""
+    agent._paused.clear()
+    msg = InternalMessage(
+        to=agent.name,
+        sender=face_module.settings.agent_settings.user_interrupt_name,
+        thread="cmd",
+        body="RESUME",
+    )
+    await agent.handle_message(msg)
+    assert agent._paused.is_set()
+
+
+@pytest.mark.asyncio
+async def test_handle_unknown_command(agent):
+    """Test that an unknown command from UserInterruptAgent is ignored (logs a warning)."""
+    msg = InternalMessage(
+        to=agent.name,
+        sender=face_module.settings.agent_settings.user_interrupt_name,
+        thread="cmd",
+        body="???",
+    )
+    await agent.handle_message(msg)
+
+
+@pytest.mark.asyncio
+async def test_handle_unknown_sender(agent):
+    """Test that messages from unknown senders are ignored."""
+    msg = InternalMessage(
+        to=agent.name,
+        sender="someone_else",
+        thread="cmd",
+        body="PAUSE",
+    )
+    await agent.handle_message(msg)

test/unit/agents/user_interrupt/test_user_interrupt.py

@@ -303,29 +303,6 @@ async def test_send_experiment_control(agent):
     assert msg.thread == "reset_experiment"
 
 
-@pytest.mark.asyncio
-async def test_send_pause_command(agent):
-    await agent._send_pause_command("true")
-    # Sends to RI and VAD
-    assert agent.send.await_count == 2
-    msgs = [call.args[0] for call in agent.send.call_args_list]
-
-    ri_msg = next(m for m in msgs if m.to == settings.agent_settings.ri_communication_name)
-    assert json.loads(ri_msg.body)["endpoint"] == ""  # PAUSE endpoint
-    assert json.loads(ri_msg.body)["data"] is True
-
-    vad_msg = next(m for m in msgs if m.to == settings.agent_settings.vad_name)
-    assert vad_msg.body == "PAUSE"
-
-    agent.send.reset_mock()
-    await agent._send_pause_command("false")
-    assert agent.send.await_count == 2
-    vad_msg = next(
-        m for m in agent.send.call_args_list if m.args[0].to == settings.agent_settings.vad_name
-    ).args[0]
-    assert vad_msg.body == "RESUME"
-
-
 @pytest.mark.asyncio
 async def test_setup(agent):
     """Test the setup method initializes sockets correctly."""