* feat(server): cross-node RSSI-weighted feature fusion + benchmarks
Adds fuse_multi_node_features() that combines CSI features across all
active ESP32 nodes using RSSI-based weighting (closer node = higher weight).
Benchmark results (2 ESP32 nodes, 30s, ~1500 frames):
Metric | Baseline | Fusion | Improvement
---------------------|----------|---------|------------
Variance mean | 109.4 | 77.6 | -29% noise
Variance std | 154.1 | 105.4 | -32% stability
Confidence | 0.643 | 0.686 | +7%
Keypoint spread std | 4.5 | 1.3 | -72% jitter
Presence ratio | 93.4% | 94.6% | +1.3pp
Person count still fluctuates near threshold — tracked as known issue.
Verified on real hardware: COM6 (node 1) + COM9 (node 2) on ruv.net.
Co-Authored-By: claude-flow <ruv@ruv.net>
* fix(ui): add client-side lerp smoothing to pose renderer
Keypoints now interpolate between frames (alpha=0.25) instead of
jumping directly to new positions. This eliminates visual jitter
that persists even with server-side EMA smoothing, because the
renderer was drawing every WebSocket frame at full rate.
Applied to skeleton, keypoints, and dense body rendering paths.
Co-Authored-By: claude-flow <ruv@ruv.net>
* feat: DynamicMinCut person separation + UI lerp smoothing
- Added ruvector-mincut dependency to sensing server
- Replaced variance-based person scoring with actual graph min-cut on
subcarrier temporal correlation matrix (Pearson correlation edges,
DynamicMinCut exact max-flow)
- Recalibrated feature scaling for real ESP32 data ranges
- UI: client-side lerp interpolation (alpha=0.25) on keypoint positions
- Dampened procedural animation (noise, stride, extremity jitter)
- Person count thresholds retuned for mincut ratio
Co-Authored-By: claude-flow <ruv@ruv.net>
* docs: update CHANGELOG with v0.5.1-v0.5.3 releases
Co-Authored-By: claude-flow <ruv@ruv.net>
* chore: update vendored ruvector to latest main (v2.1.0-40)
Was at v2.0.5-172 (f8f2c600a), now at v2.1.0-40 (050c3fe6f).
316 commits with new crates: ruvector-coherence, sona, ruvector-core,
ruvector-gnn improvements, and security hardening.
Co-Authored-By: claude-flow <ruv@ruv.net>
* feat: RuVector Phases 2+3 — temporal smoothing, kinematic constraints, coherence gating
Phase 2 (sensing server):
- Temporal keypoint smoothing via EMA (alpha=0.3) with coherence-adaptive blending
- Coherence scoring: running variance of motion_energy over 20 frames
- Low coherence → reduce alpha to 0.1 (trust measurements less)
- Per-node prev_keypoints for frame-to-frame smoothing
- Bone length clamping (±20%) in derive_single_person_pose
Phase 3 (signal crate):
- SkeletonConstraints: Jakobsen relaxation (3 iterations) on 12-bone
COCO-17 kinematic tree — prevents impossible skeletons
- CompressedPoseHistory: two-tier storage (hot f32 + warm i16 quantized)
for trajectory matching and re-ID
- 8 new tests for constraints + history
Vendored ruvector updated to v2.1.0-40 (latest main, 316 commits).
Workspace deps remain at v2.0.4 (crates.io) until v2.1.0 is published.
647 tests pass across both crates (0 failures).
Refs #296
Co-Authored-By: claude-flow <ruv@ruv.net>
* fix(server): use max instead of sum for multi-node person aggregation
With nodes in the same room, each node sees the same people. Summing
per-node counts double-counted (2 nodes × 1 person = 2 persons).
Now uses max() so 2 nodes × 1 person = 1 person.
Verified on real hardware: COM6 (node 1) + COM9 (node 2) on ruv.net,
estimated_persons=1 with 1 person in room.
Co-Authored-By: claude-flow <ruv@ruv.net>
* fix(server): reduce skeleton jitter + raise person count thresholds
- EMA alpha 0.3→0.15, low-coherence 0.1→0.05
- Remove tick-based noise (main jitter source)
- Breathing 5x slower, extremity jitter 3x smaller, stride 2x smaller
- Person count 1→2 threshold 0.65→0.80
- Aggregation sum→max for same-room nodes
Verified on COM6+COM9: 1 person stable.
Co-Authored-By: claude-flow <ruv@ruv.net>
* feat(signal): subcarrier importance weighting via mincut partition (Phase 1)
Adds subcarrier_importance_weights() to ruvector signal crate — converts
mincut partition into per-subcarrier float weights (>1.0 for sensitive,
0.5 for insensitive subcarriers).
Sensing server now uses weighted mean/variance in extract_features_from_frame
instead of treating all 56 subcarriers equally. This emphasizes body-motion-
sensitive subcarriers and reduces noise from static multipath.
Expected: ~26% reduction in keypoint jitter (±15cm → ±11cm RMS).
284 tests pass (191 trainer + 51 lib + 18 vital_signs + 16 dataset + 8 multi_node).
Co-Authored-By: claude-flow <ruv@ruv.net>
* fix(firmware): stack overflow risk + tick-rate independence (review findings)
Critical fixes from deep review:
1. **Stack overflow prevention**: Moved BPM scratch buffers (br_buf, hr_buf)
from stack to static storage in both process_frame() and
update_multi_person_vitals(). Combined stack was ~6.5-7.5 KB of 8 KB
limit — now reduced by ~4 KB to safe margins.
2. **Tick-rate independence**: Post-batch yield now uses
pdMS_TO_TICKS(20) with min-1 guard instead of raw vTaskDelay(2).
Previously assumed 100Hz tick rate.
3. **EDGE_BATCH_LIMIT to header**: Moved from local const to
edge_processing.h #define for configurability.
Firmware builds clean at 843 KB.
Co-Authored-By: claude-flow <ruv@ruv.net>
* fix(server): stale node eviction, remove unsafe pointer (review findings)
Critical fixes from deep review:
1. **Stale node eviction**: node_states HashMap now evicts nodes with no
frame for >60 seconds, every 100 ticks. Prevents unbounded memory
growth and stale smoothing data when nodes are replaced.
2. **Remove unsafe raw pointer**: Replaced the unsafe raw pointer to
adaptive_model (used to break borrow checker deadlock with
node_states) with a safe .clone() before the mutable borrow.
AdaptiveModel derives Clone so this is a clean copy.
284 tests pass, zero failures.
Co-Authored-By: claude-flow <ruv@ruv.net>
The server parsed rssi from buf[14] and noise_floor from buf[15], but
the firmware (csi_collector.c) packs them at buf[16] and buf[17]:
Firmware: n_subcarriers=u16(6-7) freq=u32(8-11) seq=u32(12-15) rssi=i8(16)
Server: n_subcarriers=u8(6) freq=u16(8-9) seq=u32(10-13) rssi=i8(14) ← WRONG
This caused RSSI to read the high byte of the sequence counter instead
of the actual signed RSSI value, producing positive values (e.g., +9)
instead of the correct negative values (e.g., -46 dBm).
Added inline documentation of the frame layout matching csi_collector.c.
Closes#332
Complements #326 (per-node state pipeline) with additional features:
- Dynamic adaptive classifier: discover activity classes from training
data filenames instead of hardcoded array. Users add classes via
filename convention (train_<class>_<desc>.jsonl), no code changes.
- Per-node UI cards: SensingTab shows individual node status with
color-coded markers, RSSI, variance, and classification per node.
- Colored node markers in 3D gaussian splat view (8-color palette).
- Per-node RSSI history tracking in sensing service.
- XSS fix: UI uses createElement/textContent instead of innerHTML.
- RSSI sign fix: ensure dBm values are always negative.
- GET /api/v1/nodes endpoint for per-node health monitoring.
- node_features field in WebSocket SensingUpdate messages.
- Firmware watchdog fix: yield after every frame to prevent IDLE1 starvation.
Addresses #237, #276, #282
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* docs(adr): ADR-068 per-node state pipeline for multi-node sensing (#249)
Documents the architectural change from single shared state to per-node
HashMap<u8, NodeState> in the sensing server. Includes scaling analysis
(256 nodes < 13 MB), QEMU validation plan, and aggregation strategy.
Also links README hero image to the explainer video.
Co-Authored-By: claude-flow <ruv@ruv.net>
* feat(server): per-node state pipeline for multi-node sensing (ADR-068, #249)
Replaces the single shared state pipeline with per-node HashMap<u8, NodeState>.
Each ESP32 node now gets independent:
- frame_history (temporal analysis)
- smoothed_person_score / prev_person_count
- smoothed_motion / baseline / debounce state
- vital sign detector + smoothing buffers
- RSSI history
Multi-node aggregation:
- Person count = sum of per-node counts for active nodes (seen <10s)
- SensingUpdate.nodes includes all active nodes
- estimated_persons reflects cross-node aggregate
Single-node deployments behave identically (HashMap has one entry).
Simulated data path unchanged for backward compatibility.
Closes#249
Refs #237, #276, #282
Co-Authored-By: claude-flow <ruv@ruv.net>
* fix(firmware,server): watchdog crash on busy LANs + no detection from edge vitals (#321, #323)
**Firmware (#321):** edge_dsp task now batch-limits frame processing to 4
frames before a 10ms yield. On corporate LANs with high CSI frame rates,
the previous 1-tick-per-frame yield wasn't enough to prevent IDLE1
starvation and task watchdog triggers.
**Sensing server (#323):** When ESP32 runs the edge DSP pipeline (Tier 2+),
it sends vitals packets (magic 0xC5110002) instead of raw CSI frames.
Previously, the server broadcast these as raw edge_vitals but never
generated a sensing_update, so the UI showed "connected" but "0 persons".
Now synthesizes a full sensing_update from vitals data including
classification, person count, and pose generation.
Closes#321Closes#323
Co-Authored-By: claude-flow <ruv@ruv.net>
* fix(firmware): address review findings — idle busy-spin and observability
- Fix pdMS_TO_TICKS(5)==0 at 100Hz causing busy-spin in idle path (use
vTaskDelay(1) instead)
- Post-batch yield now 2 ticks (20ms) for genuinely longer pause
- Add s_ring_drops counter to ring_push for diagnosing frame drops
- Expose drop count in periodic vitals log line
Co-Authored-By: claude-flow <ruv@ruv.net>
* fix(server): set breathing_band_power for skeleton animation from vitals
When presence is detected via edge vitals, set breathing_band_power to
0.5 so the UI's torso breathing animation works. Previously hardcoded
to 0.0 which made the skeleton appear static even when breathing rate
was being reported.
Co-Authored-By: claude-flow <ruv@ruv.net>
The source field was set to "esp32" on the first UDP frame but never
reverted when frames stopped arriving. This caused the UI to show
"Real hardware connected" indefinitely after powering off all nodes.
Changes:
- Add last_esp32_frame timestamp to AppStateInner
- Add effective_source() method with 5-second timeout
- Source becomes "esp32:offline" when no frames received within 5s
- Health endpoint shows "degraded" instead of "healthy" when offline
- All 6 status/health/info API endpoints use effective_source()
Fixes#297
Co-authored-by: Reuven <cohen@ruv-mac-mini.local>
The person-count heuristic was causing widespread flickering (#237, #249,
#280, #292) because:
1. Threshold 0.50 for 2-persons was too low — multipath reflections in
small rooms easily exceeded it
2. No actual hysteresis despite the comment claiming asymmetric thresholds
3. EMA smoothing (α=0.15) was too responsive to transient spikes
Changes:
- Raise up-thresholds: 1→2 persons at 0.65 (was 0.50), 2→3 at 0.85 (was 0.80)
- Add true hysteresis with asymmetric down-thresholds: 2→1 at 0.45, 3→2 at 0.70
- Track prev_person_count in SensingState for state-aware transitions
- Increase EMA smoothing to α=0.10 (~2s time constant at 20 Hz)
- Update all 4 call sites (ESP32, Windows WiFi, multi-BSSID, simulated)
Fixes#292, #280, #237
Co-authored-by: Reuven <cohen@ruv-mac-mini.local>
Add environment-tuned activity classification that learns from labeled
ESP32 CSI recordings, replacing brittle static thresholds.
- Adaptive classifier: 15-feature logistic regression trained from JSONL
recordings (variance, motion band, subcarrier stats: skew, kurtosis,
entropy, IQR). Trains in <1s, persists as JSON, auto-loads on restart.
- Three-stage signal smoothing: adaptive baseline subtraction (α=0.003),
EMA + trimmed-mean median filter (21-frame window), hysteresis debounce
(4 frames). Motion classification now stable across seconds, not frames.
- Vital signs stabilization: outlier rejection (±8 BPM HR, ±2 BPM BR),
trimmed mean, dead-band (±2 BPM HR), EMA α=0.02. HR holds steady for
10+ seconds instead of jumping 50 BPM every frame.
- Observatory auto-detect: always probes /health on startup, connects
WebSocket to live ESP32 data automatically.
- New API endpoints: POST /api/v1/adaptive/train, GET /adaptive/status,
POST /adaptive/unload for runtime model management.
- Updated user guide with Observatory, adaptive classifier tutorial,
signal smoothing docs, and new troubleshooting entries.
The web UI had persistent 404 errors on model, recording, and training
endpoints, and the sensing WebSocket never connected on Dashboard/Live
Demo tabs because sensingService.start() was only called lazily on
Sensing tab visit.
Server (main.rs):
- Add 14 fully-functional Axum handlers: model CRUD (7), recording
lifecycle (4), training control (3)
- Scan data/models/ and data/recordings/ at startup
- Recording writes CSI frames to .jsonl via tokio background task
- Model load/unload lifecycle with state tracking
Web UI (app.js):
- Import and start sensingService early in initializeServices() so
Dashboard and Live Demo tabs connect to /ws/sensing immediately
Mobile (ws.service.ts):
- Fix WebSocket URL builder to use same-origin port instead of
hardcoded port 3001
Mobile (jest.config.js):
- Fix testPathIgnorePatterns that was ignoring the entire test directory
Mobile (25 test files):
- Replace all it.todo() placeholder tests with real implementations
covering components, services, stores, hooks, screens, and utils
ADR-043 documents all changes.
* feat: RVF training pipeline & UI integration (ADR-036)
Implement full model training, management, and inference pipeline:
Backend (Rust):
- recording.rs: CSI recording API (start/stop/list/download/delete)
- model_manager.rs: RVF model loading, LoRA profile switching, model library
- training_api.rs: Training API with WebSocket progress streaming, simulated
training mode with realistic loss curves, auto-RVF export on completion
- main.rs: Wire new modules, recording hooks in all CSI paths, data dirs
UI (new components):
- ModelPanel.js: Dark-mode model library with load/unload, LoRA dropdown
- TrainingPanel.js: Recording controls, training config, live Canvas charts
- model.service.js: Model REST API client with events
- training.service.js: Training + recording API client with WebSocket progress
UI (enhancements):
- LiveDemoTab: Model selector, LoRA profile switcher, A/B split view toggle,
training quick-panel with 60s recording shortcut
- SettingsPanel: Full dark mode conversion (issue #92), model configuration
(device, threads, auto-load), training configuration (epochs, LR, patience)
- PoseDetectionCanvas: 10-frame pose trail with ghost keypoints and motion
trajectory lines, cyan trail toggle button
- pose.service.js: Model-inference confidence thresholds
UI (plumbing):
- index.html: Training tab (8th tab)
- app.js: Panel initialization and tab routing
- style.css: ~250 lines of training/model panel dark-mode styles
191 Rust tests pass, 0 failures. Closes#92.
Refs: ADR-036, #93
Co-Authored-By: claude-flow <ruv@ruv.net>
* fix: real RuVector training pipeline + UI service fixes
Training pipeline (training_api.rs):
- Replace simulated training with real signal-based training loop
- Load actual CSI data from .csi.jsonl recordings or live frame history
- Extract 180 features per frame: subcarrier amplitudes, temporal variance,
Goertzel frequency analysis (9 bands), motion gradients, global stats
- Train calibrated linear CSI-to-pose mapping via mini-batch gradient descent
with L2 regularization (ridge regression), Xavier init, cosine LR decay
- Self-supervised: teacher targets from derive_pose_from_sensing() heuristics
- Real validation metrics: MSE and PCK@0.2 on 80/20 train/val split
- Export trained .rvf with real weights, feature normalization stats, witness
- Add infer_pose_from_model() for live inference from trained model
- 16 new tests covering features, training, inference, serialization
UI fixes:
- Fix double-URL bug in model.service.js and training.service.js
(buildApiUrl was called twice — once in service, once in apiService)
- Fix route paths to match Rust backend (/api/v1/train/*, /api/v1/recording/*)
- Fix request body formats (session_name, nested config object)
- Fix top-level await in LiveDemoTab.js blocking module graph
- Dynamic imports for ModelPanel/TrainingPanel in app.js
- Center nav tabs with flex-wrap for 8-tab layout
Co-Authored-By: claude-flow <ruv@ruv.net>
* fix: WebSocket onOpen race condition, data source indicators, auto-start pose detection
- Fix WebSocket onOpen race condition in websocket.service.js where
setupEventHandlers replaced onopen after socket was already open,
preventing pose service from receiving connection signal
- Add 4-state data source indicator (LIVE/SIMULATED/RECONNECTING/OFFLINE)
across Dashboard, Sensing, and Live Demo tabs via sensing.service.js
- Add hot-plug ESP32 auto-detection in sensing server (auto mode runs
both UDP listener and simulation, switches on ESP32_TIMEOUT)
- Auto-start pose detection when backend is reachable
- Hide duplicate PoseDetectionCanvas controls when enableControls=false
- Add standalone Demo button in LiveDemoTab for offline animated demo
- Add data source banner and status styling
Co-Authored-By: claude-flow <ruv@ruv.net>
- Docker default changed from --source simulated to --source auto
(auto-detects ESP32 on UDP 5005, falls back to simulation)
- Pose derivation now driven by real sensing features: motion_band_power,
breathing_band_power, variance, dominant_freq_hz, change_points
- Temporal feature extraction: 100-frame circular buffer, Goertzel
breathing rate estimation (0.1-0.5 Hz), frame-to-frame L2 motion
detection, SNR-based signal quality metric
- Signal field driven by subcarrier variance spatial mapping instead
of fixed animation circle
- UI data source indicators: LIVE/RECONNECTING/SIMULATED banner on
sensing tab, estimation mode badge on live demo tab
- Setup guide panel explaining ESP32 count requirements for each
capability level (1x: presence, 3x: localization, 4x+: full pose)
- Tick rate improved from 500ms to 100ms (2fps to 10fps)
- Fixed Option<f64> division bug from PR #83
- ADR-035 documents all decisions
Closes#86
Co-Authored-By: claude-flow <ruv@ruv.net>
The UI had hardcoded localhost:8080 for HTTP and localhost:8765 for
WebSocket, causing "Backend unavailable" when served from Docker
(port 3000) or any non-default port.
Changes:
- api.config.js: BASE_URL now uses window.location.origin instead
of hardcoded localhost:8080
- api.config.js: buildWsUrl() uses window.location.host instead of
hardcoded localhost:8080
- sensing.service.js: WebSocket URL derived from page origin instead
of hardcoded localhost:8765
- main.rs: Added /ws/sensing route to the HTTP server so WebSocket
and REST are reachable on a single port
Fixes#55
Co-Authored-By: claude-flow <ruv@ruv.net>
- Snapshot best-epoch weights during training and restore before
checkpoint/RVF export (prevents exporting overfit final-epoch params)
- Add CsiToPoseTransformer::zeros() for fast zero-init when weights
will be overwritten, avoiding wasteful Xavier init during gradient
estimation (~2*param_count transformer constructions per batch)
- Deduplicate synthetic data generation in main.rs training mode
Co-Authored-By: claude-flow <ruv@ruv.net>
- Add docker/ folder with Dockerfile.rust (132MB), Dockerfile.python (569MB),
and docker-compose.yml
- Remove stale root-level Dockerfile and docker-compose files
- Implement --export-rvf CLI flag for standalone RVF package generation
- Generate wifi-densepose-v1.rvf (13KB) with model weights, vital config,
SONA profile, and training provenance
- Update README with Docker pull/run commands and RVF export instructions
- Update test count to 542+ and fix Docker port mappings
- Reply to issues #43, #44, #45 with Docker/RVF availability
Co-Authored-By: claude-flow <ruv@ruv.net>
Replace Python FastAPI + WebSocket servers with a single 2.1MB Rust binary
(wifi-densepose-sensing-server) that serves all UI endpoints:
- REST: /health/*, /api/v1/info, /api/v1/pose/current, /api/v1/pose/stats,
/api/v1/pose/zones/summary, /api/v1/stream/status
- WebSocket: /api/v1/stream/pose (pose_data with 17 COCO keypoints),
/ws/sensing (raw sensing_update stream on port 8765)
- Static: /ui/* with no-cache headers
WiFi-derived pose estimation: derive_pose_from_sensing() generates 17 COCO
keypoints from CSI/WiFi signal data with motion-driven animation.
Data sources: ESP32 CSI via UDP :5005, Windows WiFi via netsh, simulation
fallback. Auto-detection probes each in order.
UI changes:
- Point all endpoints to Rust server on :8080 (was Python :8000)
- Fix WebSocket sensing URL to include /ws/sensing path
- Remove sensingOnlyMode gating — all tabs init normally
- Remove api.service.js sensing-only short-circuit
- Fix clearPingInterval bug in websocket.service.js
Also removes obsolete k8s/ template manifests.
Co-Authored-By: claude-flow <ruv@ruv.net>
ruv