infinicode 2.8.89 → 2.8.91

package/dist/kernel/federation/dashboard-html.js

@@ -4004,7 +4004,7 @@ button.act:disabled{opacity:0.5;cursor:default;}
     var _statusMem = {};   // name -> { lastActiveAt, lastEndedAt, prevReadiness }
     var drawerId=null, camRAF=null, meterRAF=null, camAbort=null, drawerTimer=null, pipelineTimer=null, deviceSelectorsReady=false;
     var previewOn=false, previewTimer=null;   // on-demand operator camera preview
-    var visionOn=false;   // RoboVisionAI_PI overlay feed (direct MJPEG, not LiveKit)
+    var visionOn=false, visionRetryTimer=null;   // RoboVisionAI_PI overlay feed (direct MJPEG, not LiveKit)
     var lkRoom=null, micAnalyser=null, micMonitorSource=null, micMonitorGain=null, audioCtx=null, lkAudioEls=[], micMonitorConnecting=false, micMonitorState='no active conversation session', micMonitorAudible=false, micMonitorVolume=0.8;
     var imgs={}; // preloaded robot concept avatars
     var TW=44, TH=22, PARK_W=15, PARK_D=8;
@@ -6217,8 +6217,8 @@ button.act:disabled{opacity:0.5;cursor:default;}
       directImg.style.display='block';
       if(directCamNote)directCamNote.textContent='opening RoboVision camera relay...';
       if(directRes)directRes.textContent='CONNECTING';
-      directImg.onerror=function(){ visionOn=false; directImg.style.display='none'; directImg.src=''; if(directCamNote) directCamNote.textContent='RoboVision feed unavailable on robot'; if(directLiveNote) directLiveNote.textContent='camera unavailable'; if(directRes) directRes.textContent='OFFLINE'; };
-      directImg.onload=function(){ directImg.style.display='block'; if(directCamNote) directCamNote.textContent='live camera via RoboVision mesh relay'; if(directLiveNote) directLiveNote.textContent='live'; if(directRes) directRes.textContent='LIVE'; };
+      directImg.onerror=function(){ visionOn=false; directImg.style.display='none'; if(directCamNote) directCamNote.textContent='RoboVision feed unavailable; reconnecting...'; if(directLiveNote) directLiveNote.textContent='camera reconnecting'; if(directRes) directRes.textContent='RETRYING'; if(visionRetryTimer)clearTimeout(visionRetryTimer);visionRetryTimer=setTimeout(function(){if(drawerId&&byId[drawerId]===n&&!visionOn)tryVisionOverlay(n,true);},3000); };
+      directImg.onload=function(){ visionOn=true;if(visionRetryTimer){clearTimeout(visionRetryTimer);visionRetryTimer=null;}directImg.style.display='block'; if(directCamNote) directCamNote.textContent='live camera via RoboVision mesh relay'; if(directLiveNote) directLiveNote.textContent='live'; if(directRes) directRes.textContent='LIVE'; };
       directImg.src='/fed/media/robots/'+encodeURIComponent(directRef)+'/video_feed'+QS+(QS?'&':'?')+'_='+Date.now();
       return;
       var img=$('pk-vision-img'); if(!img) return;
@@ -6235,9 +6235,10 @@ button.act:disabled{opacity:0.5;cursor:default;}
       img.onload=function(){ img.style.display='block'; };
       img.src='http://'+ip+':'+port+'/video_feed';
     }
-    function stopVision(){
-      visionOn=false;
-      var img=$('pk-vision-img'); if(img){ img.src=''; img.style.display='none'; }
+    function stopVision(){
+      visionOn=false;
+      if(visionRetryTimer){clearTimeout(visionRetryTimer);visionRetryTimer=null;}
+      var img=$('pk-vision-img'); if(img){ img.src=''; img.style.display='none'; }
     }
     function pollRoboparkQuiet(){
       fetch('/robopark/api/telemetry'+QS,{cache:'no-store'}).then(function(r){return r.json();}).then(function(d){
@@ -6874,7 +6875,7 @@ button.act:disabled{opacity:0.5;cursor:default;}
     function set(id,v){ var e=$(id); if(e) e.textContent=v; }
     function setFill(id,pct){ var e=$(id); if(!e||pct==null) return; e.style.width=Math.max(2,Math.min(100,pct))+'%'; e.style.background=pct>=75?'var(--bad)':pct>=45?'var(--warn)':'var(--gold-bright)'; }
 
-    var cameraOpsNode=null,cameraOpsTimer=null;
+    var cameraOpsNode=null,cameraOpsTimer=null,cameraOpsRetryTimer=null;
     function cameraOpsFeedUrl(n){
       var ref=n&&(n.displayName||n.name||n.nodeId||n.id); if(!ref)return '';
       return '/fed/media/robots/'+encodeURIComponent(ref)+'/video_feed'+QS+(QS?'&':'?')+'_='+Date.now();
@@ -6892,13 +6893,13 @@ button.act:disabled{opacity:0.5;cursor:default;}
     }
     function reloadCameraOperations(){
       if(!cameraOpsNode)return;var img=$('pk-camera-ops-img'),empty=$('pk-camera-ops-empty');if(!img)return;img.dataset.live='false';if(empty){empty.style.display='block';empty.textContent='opening RoboVision camera...';}
-      img.onload=function(){img.dataset.live='true';if(empty)empty.style.display='none';refreshCameraOperations();};img.onerror=function(){img.dataset.live='false';if(empty){empty.style.display='block';empty.textContent='camera relay unavailable / retrying';}refreshCameraOperations();};img.src=cameraOpsFeedUrl(cameraOpsNode);refreshCameraOperations();
+      img.onload=function(){img.dataset.live='true';if(cameraOpsRetryTimer){clearTimeout(cameraOpsRetryTimer);cameraOpsRetryTimer=null;}if(empty)empty.style.display='none';refreshCameraOperations();};img.onerror=function(){var failedNode=cameraOpsNode;img.dataset.live='false';if(empty){empty.style.display='block';empty.textContent='camera relay unavailable / retrying';}refreshCameraOperations();if(cameraOpsRetryTimer)clearTimeout(cameraOpsRetryTimer);cameraOpsRetryTimer=setTimeout(function(){if(cameraOpsNode&&cameraOpsNode===failedNode)reloadCameraOperations();},3000);};img.src=cameraOpsFeedUrl(cameraOpsNode);refreshCameraOperations();
     }
     function openCameraOperations(n){
       cameraOpsNode=n;var host=$('pk-camera-ops');if(!host)return;host.classList.add('open');document.body.classList.add('camera-ops-open');var name=n.displayName||n.name||'robot',avatar=rpAvatarFor(name),av=$('pk-camera-ops-avatar');$('pk-camera-ops-name').textContent=name+' / live operations';$('pk-camera-ops-sub').textContent='RoboVision mesh relay / '+(n.connected?'connected':'offline');if(av){if(avatar){av.src=avatar+QS;av.style.display='block';}else{av.removeAttribute('src');av.style.display='none';}}reloadCameraOperations();if(cameraOpsTimer)clearInterval(cameraOpsTimer);cameraOpsTimer=setInterval(refreshCameraOperations,1000);
     }
     function closeCameraOperations(){
-      var host=$('pk-camera-ops'),img=$('pk-camera-ops-img');if(host)host.classList.remove('open');document.body.classList.remove('camera-ops-open');if(img){img.src='';img.dataset.live='false';}if(cameraOpsTimer){clearInterval(cameraOpsTimer);cameraOpsTimer=null;}cameraOpsNode=null;if(document.fullscreenElement&&document.exitFullscreen)document.exitFullscreen().catch(function(){});
+      var host=$('pk-camera-ops'),img=$('pk-camera-ops-img');if(host)host.classList.remove('open');document.body.classList.remove('camera-ops-open');if(cameraOpsRetryTimer){clearTimeout(cameraOpsRetryTimer);cameraOpsRetryTimer=null;}if(img){img.src='';img.dataset.live='false';}if(cameraOpsTimer){clearInterval(cameraOpsTimer);cameraOpsTimer=null;}cameraOpsNode=null;if(document.fullscreenElement&&document.exitFullscreen)document.exitFullscreen().catch(function(){});
     }
     (function bindCameraOperations(){
       var close=$('pk-camera-ops-close'),refresh=$('pk-camera-ops-refresh'),nativeBtn=$('pk-camera-ops-native'),trigger=$('pk-camera-ops-trigger'),prod=$('pk-camera-ops-production'),end=$('pk-camera-ops-end');

package/dist/kernel/federation/transport-http.js

@@ -85,7 +85,15 @@ export class MeshServer {
     }
     async streamHttp(req, res, target, headers = {}) {
         const ctrl = new AbortController();
-        req.once('close', () => ctrl.abort());
+        let clientClosed = false;
+        let timeout;
+        const armTimeout = () => {
+            if (timeout)
+                clearTimeout(timeout);
+            timeout = setTimeout(() => ctrl.abort('camera upstream timed out'), 10_000);
+        };
+        req.once('close', () => { clientClosed = true; ctrl.abort(); });
+        armTimeout();
         try {
             const upstream = await fetch(target, { headers, signal: ctrl.signal });
             if (!upstream.ok || !upstream.body) {
@@ -105,6 +113,7 @@ export class MeshServer {
                 const { done, value } = await reader.read();
                 if (done || res.writableEnded)
                     break;
+                armTimeout();
                 if (!res.write(Buffer.from(value)))
                     await once(res, 'drain');
             }
@@ -112,16 +121,20 @@ export class MeshServer {
                 res.end();
         }
         catch (err) {
-            if (ctrl.signal.aborted)
+            if (clientClosed)
                 return;
             if (!res.headersSent) {
                 res.writeHead(502, { 'content-type': 'text/plain; charset=utf-8', 'cache-control': 'no-store' });
-                res.end(`robot camera unavailable: ${err instanceof Error ? err.message : String(err)}`);
+                res.end(`robot camera unavailable: ${ctrl.signal.aborted ? 'upstream frame timeout' : (err instanceof Error ? err.message : String(err))}`);
             }
             else if (!res.writableEnded) {
                 res.end();
             }
         }
+        finally {
+            if (timeout)
+                clearTimeout(timeout);
+        }
     }
     streamLocalRoboVision(req, res) {
         void this.streamHttp(req, res, 'http://127.0.0.1:5000/video_feed');

package/docs/ROBOPARK_PRODUCTION_MEMORY.md

@@ -82,6 +82,12 @@ motion detection consume a shared latest-frame stream. Starting one
 `VideoCapture.read()` calls, which can block the second client and make the hub
 relay time out with zero bytes.
 
+Long-running camera operation also requires recovery at every layer: the Pi
+releases and reopens a V4L2 handle after an eight-second blocked read, the mesh
+relay returns an error after ten seconds without an upstream frame, and both
+drawer and fullscreen camera views reconnect every three seconds. A camera
+request must never remain in an unbounded `connecting` state.
+
 ### Duplicate Speaker Tests
 
 Only one robot process may claim a scheduler `speaker_test`. The scheduler

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "infinicode",
-  "version": "2.8.89",
+  "version": "2.8.91",
   "description": "OpenKernel — provider-agnostic AI execution kernel. Native coding agent + mission-driven execution runtime.",
   "type": "module",
   "main": "./dist/kernel/index.js",

package/packages/robopark/scheduler/preview_agent.py

@@ -455,6 +455,9 @@ class PreviewAgent:
         self._vision_session_id: Optional[str] = None
         self._vision_trigger_in_flight = False
         self._speaker_test_in_flight = False
+        # The USB adapter is an exclusive ALSA endpoint. Keep operator tests
+        # and motion cues from racing each other while LiveKit is paused.
+        self._speaker_operation_lock = asyncio.Lock()
         self._vision_hard_deadline: float = 0.0
         self._vision_last_activity: float = 0.0
         self.production_mode = False
@@ -896,18 +899,20 @@ class PreviewAgent:
                 return
             self._speaker_test_in_flight = True
             from robot_supervisor import _speaker_roundtrip_test
-            # LiveKit owns the USB microphone continuously. Release that owner
-            # for an explicit hardware test, then restore the same session.
-            restore_state = self._current_state if self._publisher else None
-            if restore_state:
-                await self._stop_publisher()
-                await asyncio.sleep(1.0)
-            try:
-                result = await asyncio.to_thread(_speaker_roundtrip_test, request.get("params") or {})
-                result["media_owner_paused"] = bool(restore_state)
-            finally:
-                if restore_state and restore_state.active:
-                    await self._start_publisher(restore_state)
+            async with self._speaker_operation_lock:
+                # LiveKit owns the USB microphone and speaker continuously.
+                # Pause it while holding the same lock as the motion cue so
+                # nothing can reopen ALSA before the explicit test starts.
+                restore_state = self._current_state if self._publisher else None
+                if restore_state:
+                    await self._stop_publisher()
+                    await asyncio.sleep(1.0)
+                try:
+                    result = await asyncio.to_thread(_speaker_roundtrip_test, request.get("params") or {})
+                    result["media_owner_paused"] = bool(restore_state)
+                finally:
+                    if restore_state and restore_state.active:
+                        await self._start_publisher(restore_state)
             result_response = await self._session.post(
                 f"{self.scheduler_url.rstrip('/')}/api/devices/{self.device_id}/supervisor-output",
                 json={"kind": "speaker_test", "service": None, "payload": result,
@@ -1066,7 +1071,8 @@ class PreviewAgent:
         # The cue and TTS share one physical ALSA output. Finish and release
         # the cue before LiveKit can deliver the greeting; background playback
         # races aplay and makes greeting delivery intermittently fail busy.
-        await asyncio.to_thread(_play_audio_effect, self.audio_playback_device, "motion")
+        async with self._speaker_operation_lock:
+            await asyncio.to_thread(_play_audio_effect, self.audio_playback_device, "motion")
         try:
             r = await self._session.post(
                 f"{self.scheduler_url.rstrip('/')}/api/devices/{self.device_id}/request-session",
@@ -1317,6 +1323,11 @@ class LiveKitPublisher:
                         process.wait(timeout=2)
                     except subprocess.TimeoutExpired:
                         process.terminate()
+                        try:
+                            process.wait(timeout=1)
+                        except subprocess.TimeoutExpired:
+                            process.kill()
+                            process.wait(timeout=1)
 
                 def close(self) -> None:
                     return
@@ -1493,16 +1504,24 @@ class LiveKitPublisher:
                 else:
                     data = b"".join(raw[i:i + 2] * OUT_CHANNELS for i in range(0, len(raw), 2))
                 write_queue.put(data)
-        except Exception as e:
-            logger.warning(f"audio out stream error: {e}")
-        finally:
-            write_queue.put(None)
-            writer_thread.join(timeout=2.0)
+        except Exception as e:
+            logger.warning(f"audio out stream error: {e}")
+        finally:
+            cancelling = bool(asyncio.current_task() and asyncio.current_task().cancelling())
+            if cancelling:
+                # On publisher teardown, close ALSA first. Waiting for the
+                # writer while aplay still owns the device leaves hw:X,Y busy
+                # long enough for the queued dashboard test to fail.
+                stream_failed.set()
+                out.stop_stream()
+            write_queue.put(None)
+            writer_thread.join(timeout=2.0)
             logger.info(
                 f"audio out: {sid} received {frame_count} frames, "
                 f"writer wrote {written_frames[0]} chunks, queue backlog at close={write_queue.qsize()}"
             )
-            out.stop_stream()
+            if not cancelling:
+                out.stop_stream()
             out.close()
             if pa is not None:
                 pa.terminate()

package/packages/robopark/vision/app_pi_clean.py

@@ -23,6 +23,8 @@ frame_condition = threading.Condition()
 latest_frame_bytes = None
 latest_frame_sequence = 0
 camera_worker_started = False
+camera_read_started_at = 0.0
+camera_last_frame_at = 0.0
 audio_input_device = "default"
 audio_output_device = "default"
 ROBOVISION_AUDIO_URL = os.getenv("ROBOVISION_AUDIO_URL", "http://127.0.0.1:8000")
@@ -233,6 +235,7 @@ def camera_worker():
     global latest_detections, motion_detection_active, motion_detected_state
     global last_motion_time, motion_frame_buffer
     global latest_frame_bytes, latest_frame_sequence
+    global camera_read_started_at, camera_last_frame_at, camera
 
     prev_gray = None
 
@@ -243,11 +246,24 @@ def camera_worker():
             time.sleep(1)
             continue
 
-        with camera_lock:
+        camera_read_started_at = time.monotonic()
+        try:
+            # This is the only camera reader. Do not hold camera_lock here:
+            # the watchdog must be able to release a wedged V4L2 handle.
             success, frame = cam.read()
+        except Exception as exc:
+            print(f"Camera read error: {exc}")
+            success, frame = False, None
+        finally:
+            camera_read_started_at = 0.0
         if not success:
+            with camera_lock:
+                if camera is cam:
+                    camera.release()
+                    camera = None
             time.sleep(0.1)
             continue
+        camera_last_frame_at = time.monotonic()
 
         # Run object detection
         processed_frame, detections = detect_objects(frame, conf_threshold=0.5)
@@ -312,6 +328,25 @@ def _ensure_camera_worker():
             return
         camera_worker_started = True
     threading.Thread(target=camera_worker, name='robovision-camera', daemon=True).start()
+    threading.Thread(target=_camera_watchdog, name='robovision-camera-watchdog', daemon=True).start()
+
+
+def _camera_watchdog():
+    global camera, camera_read_started_at
+    while True:
+        time.sleep(2.0)
+        started = camera_read_started_at
+        if not started or time.monotonic() - started < 8.0:
+            continue
+        print("Camera read stalled for 8s; releasing V4L2 handle")
+        with camera_lock:
+            if camera is not None:
+                try:
+                    camera.release()
+                except Exception:
+                    pass
+                camera = None
+        camera_read_started_at = 0.0
 
 
 def generate_frames():
@@ -339,10 +374,23 @@ def index():
 def video_feed():
     return Response(generate_frames(), mimetype='multipart/x-mixed-replace; boundary=frame')
 
-@app.route('/api/detections')
-def get_detections():
-    with lock:
-        return jsonify(latest_detections)
+@app.route('/api/detections')
+def get_detections():
+    with lock:
+        return jsonify(latest_detections)
+
+
+@app.route('/api/camera/status')
+def camera_status():
+    age = None if not camera_last_frame_at else round(time.monotonic() - camera_last_frame_at, 3)
+    return jsonify({
+        "device": str(current_camera_index),
+        "open": bool(camera is not None and camera.isOpened()),
+        "worker_started": camera_worker_started,
+        "frame_sequence": latest_frame_sequence,
+        "last_frame_age_seconds": age,
+        "read_stalled": bool(camera_read_started_at and time.monotonic() - camera_read_started_at >= 8.0),
+    })
 
 @app.route('/api/caption')
 def get_caption():