RuView NLOSConsumer NLOS perception
See what direct sight cannot.
RuView combines motion induced LiDAR aperture sampling, bounded temporal inference, and governed RF context to reveal movement around blind corners without cameras, giving people and machines earlier spatial awareness.
Interactive concept modelG0 synthetic evidence
Hidden target92%narrower synthetic uncertainty
Why this matters
Know what is happening before it enters view.
RuView extends spatial awareness into blind and occluded areas. The near term value is earlier, privacy preserving context for navigation, occupancy, and controlled safety research.
Safer blind corner navigation
Give robots an early hidden target hypothesis before committing to a turn or doorway.
Benefit · more reaction time and fewer occlusion driven stopsPrivacy first occupancy
Understand movement around entrances and interior corners without recording identifiable imagery.
Benefit · spatial context with a smaller privacy footprintBlind zone awareness
Provide advisory context near shelving, machinery, corridors, and loading areas where direct view is unreliable.
Benefit · earlier human and object awarenessOcclusion resilient presence
Study identity free movement continuity in care environments without positioning cameras in private spaces.
Benefit · persistent research signals with explicit evidence limitsAvailable now
Secure clients, synthetic tracking, governed transport, temporal memory, and the hardware test protocol.
Still gatedPhysical around the corner reproduction and measured CSI improvement.
Evidence boundary
Implemented is not the same as physically proven.
The architecture, contracts, clients, simulation, security controls, tests, and CI are built. Apple raw transient access, physical 27 fps reproduction, and measured CSI improvement remain unverified. Native Swift tests and an unsigned iOS Simulator build now pass on macOS CI.
Perception pipeline
Transient input
Near raw time of flight histograms
Motion aperture
Pose aligned temporal accumulation
NLOS inversion
Hidden target hypotheses
RuView memory
Tracks, confidence, provenance
System architecture
Weak signals become governed spatial memory.
Every stage preserves evidence level, provenance, confidence, expiry, and rejection reasons. The system refuses to turn simulation into a hardware claim.
Synthetic aperture lab
Move the sensor. Watch uncertainty collapse.
This model explains the mechanism and expected direction of change. It does not reproduce the MIT algorithm or generate physical evidence.
Interactive synthetic modelNot hardware evidence
Move the sensor and accumulate frames. More viewpoint diversity increases the virtual aperture. RF contributes a coarse prior, not optical geometry.
Virtual aperture quality47%
Practical capability
Optical precision meets RF persistence.
The business value is privacy preserving spatial awareness in environments where cameras are undesirable, blocked, or fragile.
Around corner presenceHighHighVery high
Through wall presenceLowHighHigh
Precise geometryHighLowHigh
Persistent sensingLowHighHigh
Occlusion resilienceMediumHighVery high
CapabilityLiDARRFFused
Implementation status
Built software. Open physical proof.
The current release is suitable for integration, simulation, contract testing, transport testing, and preparing a controlled hardware trial.
Review source and validation in PR 1687Rust sensing layer
Transient contracts, calibration, tracking, transport, bounded readers, and world model bridge.
36 tests passedWeb and iOS clients
Expo web and iOS bundles plus native Swift contracts, secure sessions, provenance, and stale state handling.
250 web tests passedNative Apple validation
Swift protocol and security tests plus the unsigned iOS Simulator build passed on macOS CI. Raw transient access remains a hardware API uncertainty.
macOS CI passedPhysical reproduction
Run the pinned MIT method with a supported sensor and witness sustained tracking at 27 fps or better.
Required before hardware claimMeasured CSI fusion
Capture paired LiDAR and RF sessions and demonstrate at least 25 percent lower error or lost track rate.
Required before fusion claimEvidence remains explicit
ARKit depth only input is rejected for live NLOS. L3 evidence and measured fusion are rejected in the version one contract.
Fail closed by designResearch lineage
Built from published physics. Extended with governed multimodal memory.
MIT demonstrated that tiny sensor motion can synthesize an aperture large enough to recover weak multipath LiDAR returns. RuView adds bounded temporal state, RF context, evidence controls, native clients, and a world model integration path.