chat: mission mode (/mission, --mission) — autonomous goal runs in background with retries, corrections, state-aware retry hints; input() moved to thread so missions run while the REPL waits; configurable nudge_limit

testbed/README.md

@@ -82,9 +82,19 @@ you> иди к маяку            → look_at + move step by step (auto-conti
 you> повернись налево       → turn(-90)
 you> осмотрись              → vision + description
 you> активируй маяк         → interact (when close)
+you> /mission               → autonomous goal: reach & activate the beacon,
+                             keeps trying until done (retries + corrections)
+you> /mission дойди до маяка
+you> /status  /stop         → mission progress / cancel
 you> /state  /look  /map  /models  /model gemma4:12b  /steps N  /help  /exit
 ```
 
+Missions run in the background while the REPL stays usable — watch the
+capsule on http://127.0.0.1:8000 as it works. Start one directly:
+`python -m testbed.chat --mission [--mission-steps 50] [--mission-retries 3]`.
+The mission keeps trying (corrections when it drifts, nudges when it stalls,
+fresh attempts on failure) until the goal is achieved; `/stop` cancels it.
+
 Each turn's transcript is printed; the current frame lands in `chat_frame.png`
 and the sensor-built map in `chat_map.png`. Any OpenAI-compatible endpoint
 works: `python -m testbed.chat --base-url https://api.openai.com/v1

testbed/chat.py

@@ -44,7 +44,12 @@ You are talking to the capsule's brain (LLM). Type what you want, e.g.:
   'повернись налево' / 'turn left'     — quick turn
   'осмотрись' / 'look around'          — vision + description
   'активируй маяк' / 'activate it'     — interact (must be close)
-Commands: /state /look /map /models /model X /steps N /help /exit"""
+Missions (autonomous goal, keeps trying until done):
+  /mission             — goal: reach and activate the beacon
+  /mission <text>      — any goal, e.g. /mission дойди до маяка
+  /status              — mission progress + capsule state
+  /stop                — cancel the mission (REPL stays usable)
+Other commands: /state /look /map /models /model X /steps N /help /exit"""
 
 
 async def chat_loop(client: AICCClient, manifest, args: argparse.Namespace) -> int:
@@ -82,9 +87,10 @@ async def chat_loop(client: AICCClient, manifest, args: argparse.Namespace) -> i
         await asyncio.to_thread(_write)
         print(f"  frame saved to chat_frame.png ({out['width']}x{out['height']})")
 
-    def save_map() -> None:
+    def save_map(quiet: bool = False) -> None:
         if controller.pos is None:
-            print("  (no position yet — ask the capsule to move first)")
+            if not quiet:
+                print("  (no position yet — ask the capsule to move first)")
             return
         from testbed.room.mapview import draw_sensor_map
 
@@ -96,18 +102,134 @@ async def chat_loop(client: AICCClient, manifest, args: argparse.Namespace) -> i
             }
         img = draw_sensor_map(layout, controller.pos, controller.yaw, controller.path)
         img.save("chat_map.png")
-        print(
-            f"  map saved to chat_map.png (pos {controller.pos[0]:.1f}, {controller.pos[1]:.1f})"
-        )
+        if not quiet:
+            print(
+                f"  map saved to chat_map.png (pos {controller.pos[0]:.1f}, {controller.pos[1]:.1f})"
+            )
 
     async def cmd_models() -> None:
         print("  fetching models…")
         for m in ollama_models(args.base_url, args.api_key):
             print(f"    {m}")
 
-    while True:
+    # -- mission: autonomous goal run in the background ---------------------
+    # The REPL stays responsive: /status to check progress, /stop to cancel.
+    mission: asyncio.Task | None = None
+    mission_goal = "reach the beacon and activate it"
+
+    def mission_log(role: str, msg: str) -> None:
+        print(f"  [{role}] {msg}")
+
+    async def map_saver() -> None:
+        while True:
+            try:
+                save_map(quiet=True)
+            except Exception as exc:  # noqa: BLE001 - never let the saver die
+                print(f"  [map] save error: {type(exc).__name__}: {exc}")
+            await asyncio.sleep(3.0)
+
+    async def mission_runner(goal_text: str) -> None:
+        from testbed.llm_agent import run_llm_agent_loop
+
+        controller.messages.append(
+            {
+                "role": "user",
+                "content": (
+                    f"MISSION (set by the user): {goal_text}. "
+                    "Keep calling tools and do not stop until the goal is achieved. "
+                    "Report only when done."
+                ),
+            }
+        )
+        saver = asyncio.create_task(map_saver())
         try:
-            text = input("you> ").strip()
+            for attempt in range(1, args.mission_retries + 2):
+                print(
+                    f"[mission] attempt {attempt}/{args.mission_retries + 1}: {goal_text}"
+                )
+                try:
+                    summary = await run_llm_agent_loop(
+                        controller,
+                        args.mission_steps,
+                        log=mission_log,
+                        nudge_limit=args.mission_nudges,
+                    )
+                except RuntimeError as exc:
+                    print(f"  [mission] LLM error: {exc}")
+                    break
+                save_map()
+                if summary["interacted"]:
+                    print(f"\n[mission] DONE: {summary['result']}")
+                    return
+                print(
+                    f"  [mission] attempt {attempt} stopped ({summary['result']}, "
+                    f"{summary['steps']} steps) — retrying"
+                )
+                controller.messages.append(
+                    {
+                        "role": "user",
+                        "content": (
+                            "You have not achieved the mission yet. "
+                            + (
+                                controller.state_hint() + " "
+                                if controller.state_hint()
+                                else ""
+                            )
+                            + "Keep trying: call tools and do not stop until the goal is achieved."
+                        ),
+                    }
+                )
+            print(
+                f"[mission] gave up after {args.mission_retries + 1} attempts — "
+                "say '/mission' to retry or command the capsule manually"
+            )
+        finally:
+            saver.cancel()
+
+    def start_mission(goal_text: str | None = None) -> None:
+        nonlocal mission, mission_goal
+        if mission is not None and not mission.done():
+            print("[mission] already running — /stop first (or wait)")
+            return
+        mission_goal = goal_text or mission_goal
+        mission = asyncio.create_task(mission_runner(mission_goal))
+
+    async def stop_mission() -> None:
+        nonlocal mission
+        if mission is not None and not mission.done():
+            mission.cancel()
+            try:
+                await mission
+            except asyncio.CancelledError:
+                pass
+            mission = None
+            print("[mission] stopped")
+        else:
+            print("  (no mission running)")
+
+    async def cmd_status() -> None:
+        if mission is not None and not mission.done():
+            print(f"[mission] running: {mission_goal}")
+        else:
+            print("  (no mission running)")
+        await cmd_state()
+
+    if args.mission:
+        start_mission()
+
+    while True:
+        # Surface a finished mission task so its result is reported once.
+        if mission is not None and mission.done():
+            try:
+                mission.result()
+            except Exception as exc:  # noqa: BLE001 - report mission failure
+                print(f"[mission] error: {type(exc).__name__}: {exc}")
+            mission = None
+            print("[mission] finished — you can start another with /mission")
+        try:
+            # input() in a thread: a blocking read here would freeze the
+            # event loop and stall a running mission task.
+            text = (await asyncio.to_thread(input, "you> ")).strip()
         except (EOFError, KeyboardInterrupt):
             print("\n[chat] bye")
             return 0
@@ -115,11 +237,29 @@ async def chat_loop(client: AICCClient, manifest, args: argparse.Namespace) -> i
             continue
         low = text.lower()
         if low in ("/exit", "exit", "quit", "выход"):
+            if mission is not None and not mission.done():
+                mission.cancel()
+                try:
+                    await mission
+                except asyncio.CancelledError:
+                    pass
             print("[chat] bye")
             return 0
         if low == "/help":
             print(HELP)
             continue
+        if low == "/mission":
+            start_mission()
+            continue
+        if low.startswith("/mission "):
+            start_mission(low.split(maxsplit=1)[1])
+            continue
+        if low == "/stop":
+            await stop_mission()
+            continue
+        if low == "/status":
+            await cmd_status()
+            continue
         if low == "/state":
             await cmd_state()
             continue
@@ -194,6 +334,7 @@ async def run(args: argparse.Namespace) -> int:
     """Run the chat session, reconnecting to the bridge if the connection dies."""
     attempts = 0
     while True:
+        session_rc: int | None = None
         try:
             transport = WebSocketClientTransport(args.url)
             async with AICCClient(transport) as client:
@@ -202,12 +343,16 @@ async def run(args: argparse.Namespace) -> int:
                     f"[handshake] session {manifest.session_id} world {manifest.world.name}"
                 )
                 print(f"[handshake] tools: {[t.id for t in manifest.tools]}")
-                return await chat_loop(client, manifest, args)
+                session_rc = await chat_loop(client, manifest, args)
         except asyncio.CancelledError:
             raise
         except KeyboardInterrupt:
             raise
         except Exception as exc:  # noqa: BLE001 - connection lost: reconnect
+            if session_rc is not None:
+                # chat_loop exited cleanly; only cleanup failed — ignore.
+                print(f"  (cleanup note: {type(exc).__name__})")
+                return session_rc
             attempts += 1
             print(f"\n[chat] connection lost ({type(exc).__name__}: {exc})")
             if attempts >= 3:
@@ -219,6 +364,8 @@ async def run(args: argparse.Namespace) -> int:
                 "[chat] reconnecting in 2 seconds… (the room keeps its state on the bridge)"
             )
             await asyncio.sleep(2.0)
+        else:
+            return session_rc if session_rc is not None else 0
 
 
 def main() -> int:
@@ -243,6 +390,29 @@ def main() -> int:
         default=8,
         help="how many tool steps the model may chain per request (0 = one action per turn)",
     )
+    parser.add_argument(
+        "--mission",
+        action="store_true",
+        help="start an autonomous mission on connect: reach and activate the beacon",
+    )
+    parser.add_argument(
+        "--mission-steps",
+        type=int,
+        default=50,
+        help="tool steps per mission attempt (default 50)",
+    )
+    parser.add_argument(
+        "--mission-retries",
+        type=int,
+        default=2,
+        help="restarts after a failed attempt (default 2)",
+    )
+    parser.add_argument(
+        "--mission-nudges",
+        type=int,
+        default=3,
+        help="how many consecutive nudges a mission may use before giving up (default 3)",
+    )
     args = parser.parse_args()
     try:
         return asyncio.run(run(args))

testbed/llm_agent.py

@@ -363,6 +363,7 @@ async def run_llm_agent_loop(
     *,
     log: LogFn,
     recorder: Any | None = None,
+    nudge_limit: int = 1,
 ) -> dict[str, Any]:
     """Drive the controller until the mission is done or steps run out."""
     summary: dict[str, Any] = {
@@ -450,11 +451,13 @@ async def run_llm_agent_loop(
         if not turn.calls:
             # Nudge a model that drifted into plain text back to acting.
             silent = (not turn.text) or len(turn.text) < 8
-            already_nudged = (
-                bool(controller.messages)
-                and controller.messages[-1].get("content") == NUDGE
-            )
-            if silent or already_nudged:
+            consecutive_nudges = 0
+            for msg in reversed(controller.messages):
+                if msg.get("content") == NUDGE:
+                    consecutive_nudges += 1
+                else:
+                    break
+            if silent or consecutive_nudges >= nudge_limit:
                 summary["result"] = "model produced no tool call"
                 return summary
             controller.messages.append({"role": "user", "content": NUDGE})