fix: default belief false, user interrupt chat role, vad initial silence, unused import

ref: N25B-453

src/control_backend/agents/bdi/bdi_program_manager.py

@@ -1,10 +1,12 @@
 import asyncio
 import json
+import logging
 
 import zmq
 from pydantic import ValidationError
 from zmq.asyncio import Context
 
+import control_backend
 from control_backend.agents import BaseAgent
 from control_backend.agents.bdi.agentspeak_generator import AgentSpeakGenerator
 from control_backend.core.config import settings
@@ -19,6 +21,8 @@ from control_backend.schemas.program import (
     Program,
 )
 
+experiment_logger = logging.getLogger(settings.logging_settings.experiment_logger_name)
+
 
 class BDIProgramManager(BaseAgent):
     """
@@ -241,6 +245,18 @@ class BDIProgramManager(BaseAgent):
         await self.send(extractor_msg)
         self.logger.debug("Sent message to extractor agent to clear history.")
 
+    @staticmethod
+    def _rollover_experiment_logs():
+        """
+        A new experiment program started; make a new experiment log file.
+        """
+        handlers = logging.getLogger("experiment").handlers
+        for handler in handlers:
+            if isinstance(handler, control_backend.logging.DatedFileHandler):
+                experiment_logger.action("Doing rollover...")
+                handler.do_rollover()
+                experiment_logger.debug("Finished rollover.")
+
     async def _receive_programs(self):
         """
         Continuous loop that receives program updates from the HTTP endpoint.
@@ -261,6 +277,7 @@ class BDIProgramManager(BaseAgent):
             self._initialize_internal_state(program)
             await self._send_program_to_user_interrupt(program)
             await self._send_clear_llm_history()
+            self._rollover_experiment_logs()
 
             await asyncio.gather(
                 self._create_agentspeak_and_send_to_bdi(program),

src/control_backend/agents/bdi/text_belief_extractor_agent.py

@@ -150,6 +150,9 @@ class TextBeliefExtractorAgent(BaseAgent):
             return
 
         available_beliefs = [b for b in belief_list.beliefs if isinstance(b, SemanticBelief)]
+        self._current_beliefs = BeliefState(
+            false={InternalBelief(name=b.name, arguments=None) for b in available_beliefs},
+        )
         self.belief_inferrer.available_beliefs = available_beliefs
         self.logger.debug(
             "Received %d semantic beliefs from the program manager: %s",
@@ -170,6 +173,9 @@ class TextBeliefExtractorAgent(BaseAgent):
         available_goals = {g for g in goals_list.goals if g.can_fail}
         available_goals -= self._force_completed_goals
         self.goal_inferrer.goals = available_goals
+        self._current_goal_completions = {
+            f"achieved_{AgentSpeakGenerator.slugify(goal)}": False for goal in available_goals
+        }
         self.logger.debug(
             "Received %d failable goals from the program manager: %s",
             len(available_goals),

src/control_backend/agents/perception/vad_agent.py

@@ -285,9 +285,10 @@ class VADAgent(BaseAgent):
                 assert self.audio_out_socket is not None
                 await self.audio_out_socket.send(self.audio_buffer[: -2 * len(chunk)].tobytes())
 
-            # At this point, we know that the speech has ended.
+            # At this point, we know that there is no speech.
-            # Prepend the last chunk that had no speech, for a more fluent boundary
+            # Prepend the last few chunks that had no speech, for a more fluent boundary.
-            self.audio_buffer = chunk
+            self.audio_buffer = np.append(self.audio_buffer, chunk)
+            self.audio_buffer = self.audio_buffer[-begin_silence_length * len(chunk) :]
 
     async def handle_message(self, msg: InternalMessage):
         """

src/control_backend/agents/user_interrupt/user_interrupt_agent.py

@@ -9,7 +9,7 @@ from control_backend.agents.bdi.agentspeak_generator import AgentSpeakGenerator
 from control_backend.core.agent_system import InternalMessage
 from control_backend.core.config import settings
 from control_backend.schemas.belief_message import Belief, BeliefMessage
-from control_backend.schemas.program import ConditionalNorm, Goal, Program
+from control_backend.schemas.program import ConditionalNorm, Program
 from control_backend.schemas.ri_message import (
     GestureCommand,
     PauseCommand,
@@ -250,18 +250,6 @@ class UserInterruptAgent(BaseAgent):
             self._cond_norm_map = {}
             self._cond_norm_reverse_map = {}
 
-            def _register_goal(goal: Goal):
-                """Recursively register goals and their subgoals."""
-                slug = AgentSpeakGenerator.slugify(goal)
-                self._goal_map[str(goal.id)] = slug
-                self._goal_reverse_map[slug] = str(goal.id)
-
-                # Recursively check steps for subgoals
-                if goal.plan and goal.plan.steps:
-                    for step in goal.plan.steps:
-                        if isinstance(step, Goal):
-                            _register_goal(step)
-
             for phase in program.phases:
                 for trigger in phase.triggers:
                     slug = AgentSpeakGenerator.slugify(trigger)
@@ -269,7 +257,8 @@ class UserInterruptAgent(BaseAgent):
                     self._trigger_reverse_map[slug] = str(trigger.id)
 
                 for goal in phase.goals:
-                    _register_goal(goal)
+                    self._goal_map[str(goal.id)] = AgentSpeakGenerator.slugify(goal)
+                    self._goal_reverse_map[AgentSpeakGenerator.slugify(goal)] = str(goal.id)
 
                 for goal, id in self._goal_reverse_map.items():
                     self.logger.debug(f"Goal mapping: UI ID {goal} -> {id}")
@@ -311,7 +300,7 @@ class UserInterruptAgent(BaseAgent):
 
         :param text_to_say: The string that the robot has to say.
         """
-        experiment_logger.chat(text_to_say, extra={"role": "user"})
+        experiment_logger.chat(text_to_say, extra={"role": "assistant"})
         cmd = SpeechCommand(data=text_to_say, is_priority=True)
         out_msg = InternalMessage(
             to=settings.agent_settings.robot_speech_name,