feat: add useful experiment logs to various agents

ref: N25B-401

src/control_backend/agents/actuation/robot_gesture_agent.py

@@ -1,4 +1,5 @@
 import json
+import logging
 
 import zmq
 import zmq.asyncio as azmq
@@ -8,6 +9,8 @@ from control_backend.core.agent_system import InternalMessage
 from control_backend.core.config import settings
 from control_backend.schemas.ri_message import GestureCommand, RIEndpoint
 
+experiment_logger = logging.getLogger(settings.logging_settings.experiment_logger_name)
+
 
 class RobotGestureAgent(BaseAgent):
     """
@@ -111,6 +114,7 @@ class RobotGestureAgent(BaseAgent):
                         gesture_command.data,
                     )
                     return
+            experiment_logger.action("Gesture: %s", gesture_command.data)
             await self.pubsocket.send_json(gesture_command.model_dump())
         except Exception:
             self.logger.exception("Error processing internal message.")

src/control_backend/agents/bdi/bdi_core_agent.py

@@ -1,6 +1,7 @@
 import asyncio
 import copy
 import json
+import logging
 import time
 from collections.abc import Iterable
 
@@ -19,6 +20,9 @@ from control_backend.schemas.ri_message import GestureCommand, RIEndpoint, Speec
 DELIMITER = ";\n"  # TODO: temporary until we support lists in AgentSpeak
 
 
+experiment_logger = logging.getLogger(settings.logging_settings.experiment_logger_name)
+
+
 class BDICoreAgent(BaseAgent):
     """
     BDI Core Agent.
@@ -207,6 +211,9 @@ class BDICoreAgent(BaseAgent):
         else:
             term = agentspeak.Literal(name)
 
+        if name != "user_said":
+            experiment_logger.observation(f"Formed new belief: {name}{f'={args}' if args else ''}")
+
         self.bdi_agent.call(
             agentspeak.Trigger.addition,
             agentspeak.GoalType.belief,
@@ -244,6 +251,9 @@ class BDICoreAgent(BaseAgent):
             new_args = (agentspeak.Literal(arg) for arg in args)
             term = agentspeak.Literal(name, new_args)
 
+        if name != "user_said":
+            experiment_logger.observation(f"Removed belief: {name}{f'={args}' if args else ''}")
+
         result = self.bdi_agent.call(
             agentspeak.Trigger.removal,
             agentspeak.GoalType.belief,
@@ -386,6 +396,8 @@ class BDICoreAgent(BaseAgent):
                 body=str(message_text),
             )
 
+            experiment_logger.chat(str(message_text), extra={"role": "assistant"})
+
             self.add_behavior(self.send(chat_history_message))
 
             yield

src/control_backend/agents/llm/llm_agent.py

@@ -1,5 +1,6 @@
 import asyncio
 import json
+import logging
 import re
 import uuid
 from collections.abc import AsyncGenerator
@@ -14,6 +15,8 @@ from control_backend.core.config import settings
 from ...schemas.llm_prompt_message import LLMPromptMessage
 from .llm_instructions import LLMInstructions
 
+experiment_logger = logging.getLogger(settings.logging_settings.experiment_logger_name)
+
 
 class LLMAgent(BaseAgent):
     """
@@ -170,7 +173,7 @@ class LLMAgent(BaseAgent):
             *self.history,
         ]
 
-        message_id = str(uuid.uuid4())  # noqa
+        message_id = str(uuid.uuid4())
 
         try:
             full_message = ""
@@ -179,10 +182,9 @@ class LLMAgent(BaseAgent):
                 full_message += token
                 current_chunk += token
 
-                self.logger.llm(
+                experiment_logger.chat(
-                    "Received token: %s",
                     full_message,
-                    extra={"reference": message_id},  # Used in the UI to update old logs
+                    extra={"role": "assistant", "reference": message_id, "partial": True},
                 )
 
                 # Stream the message in chunks separated by punctuation.
@@ -197,6 +199,11 @@ class LLMAgent(BaseAgent):
             # Yield any remaining tail
             if current_chunk:
                 yield current_chunk
+
+            experiment_logger.chat(
+                full_message,
+                extra={"role": "assistant", "reference": message_id, "partial": False},
+            )
         except httpx.HTTPError as err:
             self.logger.error("HTTP error.", exc_info=err)
             yield "LLM service unavailable."

src/control_backend/agents/perception/transcription_agent/transcription_agent.py

@@ -1,4 +1,5 @@
 import asyncio
+import logging
 
 import numpy as np
 import zmq
@@ -10,6 +11,8 @@ from control_backend.core.config import settings
 
 from .speech_recognizer import SpeechRecognizer
 
+experiment_logger = logging.getLogger(settings.logging_settings.experiment_logger_name)
+
 
 class TranscriptionAgent(BaseAgent):
     """
@@ -25,6 +28,8 @@ class TranscriptionAgent(BaseAgent):
     :ivar audio_in_socket: The ZMQ SUB socket instance.
     :ivar speech_recognizer: The speech recognition engine instance.
     :ivar _concurrency: Semaphore to limit concurrent transcriptions.
+    :ivar _current_speech_reference: The reference of the current user utterance, for synchronising
+        experiment logs.
     """
 
     def __init__(self, audio_in_address: str):
@@ -39,6 +44,7 @@ class TranscriptionAgent(BaseAgent):
         self.audio_in_socket: azmq.Socket | None = None
         self.speech_recognizer = None
         self._concurrency = None
+        self._current_speech_reference: str | None = None
 
     async def setup(self):
         """
@@ -63,6 +69,10 @@ class TranscriptionAgent(BaseAgent):
 
         self.logger.info("Finished setting up %s", self.name)
 
+    async def handle_message(self, msg: InternalMessage):
+        if msg.thread == "voice_activity":
+            self._current_speech_reference = msg.body
+
     async def stop(self):
         """
         Stop the agent and close sockets.
@@ -96,20 +106,21 @@ class TranscriptionAgent(BaseAgent):
 
     async def _share_transcription(self, transcription: str):
         """
-        Share a transcription to the other agents that depend on it.
+        Share a transcription to the other agents that depend on it, and to experiment logs.
 
         Currently sends to:
         - :attr:`settings.agent_settings.text_belief_extractor_name`
+        - The UI via the experiment logger
 
         :param transcription: The transcribed text.
         """
-        receiver_names = [
+        experiment_logger.chat(
-            settings.agent_settings.text_belief_extractor_name,
+            transcription,
-        ]
+            extra={"role": "user", "reference": self._current_speech_reference, "partial": False},
+        )
 
-        for receiver_name in receiver_names:
         message = InternalMessage(
-                to=receiver_name,
+            to=settings.agent_settings.text_belief_extractor_name,
             sender=self.name,
             body=transcription,
         )
@@ -129,10 +140,9 @@ class TranscriptionAgent(BaseAgent):
                 audio = np.frombuffer(audio_data, dtype=np.float32)
                 speech = await self._transcribe(audio)
                 if not speech:
-                    self.logger.info("Nothing transcribed.")
+                    self.logger.debug("Nothing transcribed.")
                     continue
 
-                self.logger.info("Transcribed speech: %s", speech)
                 await self._share_transcription(speech)
             except Exception as e:
                 self.logger.error(f"Error in transcription loop: {e}")

src/control_backend/agents/perception/vad_agent.py

@@ -1,4 +1,6 @@
 import asyncio
+import logging
+import uuid
 
 import numpy as np
 import torch
@@ -12,6 +14,8 @@ from control_backend.schemas.internal_message import InternalMessage
 from ...schemas.program_status import PROGRAM_STATUS, ProgramStatus
 from .transcription_agent.transcription_agent import TranscriptionAgent
 
+experiment_logger = logging.getLogger(settings.logging_settings.experiment_logger_name)
+
 
 class SocketPoller[T]:
     """
@@ -252,6 +256,18 @@ class VADAgent(BaseAgent):
             if prob > prob_threshold:
                 if self.i_since_speech > non_speech_patience + begin_silence_length:
                     self.logger.debug("Speech started.")
+                    reference = str(uuid.uuid4())
+                    experiment_logger.chat(
+                        "...",
+                        extra={"role": "user", "reference": reference, "partial": True},
+                    )
+                    await self.send(
+                        InternalMessage(
+                            to=settings.agent_settings.transcription_name,
+                            body=reference,
+                            thread="voice_activity",
+                        )
+                    )
                 self.audio_buffer = np.append(self.audio_buffer, chunk)
                 self.i_since_speech = 0
                 continue

src/control_backend/agents/user_interrupt/user_interrupt_agent.py

@@ -1,4 +1,5 @@
 import json
+import logging
 
 import zmq
 from zmq.asyncio import Context
@@ -16,6 +17,8 @@ from control_backend.schemas.ri_message import (
     SpeechCommand,
 )
 
+experiment_logger = logging.getLogger(settings.logging_settings.experiment_logger_name)
+
 
 class UserInterruptAgent(BaseAgent):
     """
@@ -296,6 +299,7 @@ class UserInterruptAgent(BaseAgent):
 
         :param text_to_say: The string that the robot has to say.
         """
+        experiment_logger.chat(text_to_say, extra={"role": "user"})
         cmd = SpeechCommand(data=text_to_say, is_priority=True)
         out_msg = InternalMessage(
             to=settings.agent_settings.robot_speech_name,