Skip to content
RFL_GLOBAL
中文
  • WIFI + LIDAR + CAM
  • THROUGH-WALL DETECTION
  • 30HZ TRACKING

OSIRIS

WE SEE THE UNSEEN

Most human localization relies on cameras, which fail in darkness, smoke, and privacy-critical environments. OSIRIS fuses WiFi RSSI fingerprinting, LiDAR point cloud processing, and RGB-D camera pose estimation into a single real-time 30Hz tracking stream. Detect and track humans through walls and in total darkness with sub-meter precision.

MODULE STATUS: R&D ACTIVE

Tracking Rate

30Hz

DIVISION
ANIMA
WAVE
W1
DOMAIN
FOUNDATION
WAVE 1 // ANIMA SUITE
MULTI-MODAL HUMAN LOCALIZATION
OSIRIS // W1 // 057/079
01THE CHALLENGE

HUMANS ARE INVISIBLE

Most human localization relies on cameras, which fail in darkness, smoke, and privacy-critical environments. GPS doesn't work indoors. LiDAR can't see through walls. WiFi signals pass through obstacles but lack spatial precision.

What if you could fuse all of them? Detect and track humans through walls and in total darkness, with sub-meter precision, in real-time across multiple floors. That's the future of emergency response, search and rescue, and smart building automation.

02THE SOLUTION

WHAT OSIRIS DELIVERS

OSIRIS is Robot Flow Labs' proprietary multi-modal human localization system. It fuses WiFi RSSI, LiDAR, and RGB-D into a single real-time tracking stream.

PIPELINE

  1. 01WiFi RSSI fingerprinting — detect human presence through walls and obstacles
  2. 02LiDAR point cloud segmentation — 3D spatial precision and human identification
  3. 03RGB-D camera pose estimation — visual tracking with depth information
  4. 04Multi-modal fusion — Kalman-Particle, Bayesian, or Transformer at 30Hz

CAPABILITIES

  • THROUGH-WALL DETECTIONWiFi RSSI signatures reveal human presence behind obstacles→ Works in darkness, smoke, and GPS-denied environments
  • SENSOR FUSIONWiFi (25%) + LiDAR (40%) + Camera (35%) weighted consensus→ Three algorithms auto-selected for current sensor availability
  • MULTI-PERSON RE-IDPersistent tracking across occlusion and sensor gaps→ Redis-backed state store for distributed multi-robot coordination
03ENGINEERING

WHY THIS IS HARD

Sensor fusion at this scale requires solving multiple hard problems simultaneously:

  1. 01WiFi RSSI fingerprinting: building radio maps robust to environmental drift and multipath
  2. 02LiDAR segmentation: distinguishing humans from furniture and clutter in 3D point clouds
  3. 03Cross-modal association: matching WiFi/LiDAR/camera detections into unified identities
  4. 04Temporal consistency: Kalman filtering and particle filtering to smooth noisy measurements
  5. 05Real-time 30Hz fusion on embedded hardware with sub-100ms latency

OSIRIS integrates Kalman-Particle, Bayesian, and Transformer fusion — each excels at different sensor combinations. The system auto-selects the best for current availability.

04BENCHMARKS

SYSTEM SPECIFICATIONS

Target performance for multi-modal fusion pipeline:

SYSTEM SPECIFICATIONS
METRICVALUE
Tracking Rate30Hz real-time
Fusion Latency<100ms on embedded hardware
Sensor ModalitiesWiFi RSSI + LiDAR + RGB-D Camera
WiFi Weight25% (presence detection)
LiDAR Weight40% (spatial precision)
Camera Weight35% (pose estimation)
Fusion AlgorithmsKalman-Particle, Bayesian, Transformer
Re-identificationMulti-person persistent tracking
05BUILD STATUS

WHAT'S BUILT TODAY

1/9 COMPONENTS COMPLETE
WHAT'S BUILT TODAY
COMPONENTSTATUSNOTES
Project ArchitectureCOMPLETESystem design and sensor interfaces defined
Core ModelsIN PROGRESSFusion engine framework under active development
Sensor AbstractionIN PROGRESSWiFi, LiDAR, Camera interface layer
Kalman-Particle FusionIN PROGRESSTemporal tracking algorithm
Bayesian FusionIN PROGRESSMulti-sensor voting algorithm
Transformer FusionIN PROGRESSCross-modal association architecture
WebSocket StreamIN PROGRESSReal-time 30Hz client updates
Redis State StoreIN PROGRESSDistributed session management
API LayerPLANNEDREST + gRPC APIs pending core completion
06APPLICATIONS

WHERE OSIRIS DEPLOYS

  • APP_01

    SEARCH & RESCUE

    Locate humans in collapsed buildings, smoke, or darkness. Through-wall detection saves lives when cameras are blind.

  • APP_02

    SMART BUILDINGS

    Detect occupancy and movement through walls for HVAC control, energy optimization, and space utilization.

  • APP_03

    PERIMETER SECURITY

    Detect human presence in GPS-denied zones. Multi-sensor fusion eliminates blind spots in critical infrastructure.

  • APP_04

    AUTONOMOUS NAVIGATION

    Mobile robots detecting humans through obstacles. Predict collisions before line-of-sight contact.

  • APP_05

    INDUSTRIAL SAFETY

    Real-time human monitoring in hazardous areas. Alert systems when personnel enter restricted zones.

  • APP_06

    PRIVACY-CRITICAL SURVEILLANCE

    Human detection without face recognition. Presence and trajectory data only — no biometric capture.

07TECHNOLOGY

UNDER THE HOOD

PROPRIETARY STACK

  • Multi-modal sensor fusion: WiFi RSSI, LiDAR point clouds, RGB-D depth cameras
  • Kalman-Particle filtering for robust temporal tracking across sensor dropouts
  • Bayesian consensus for multi-sensor voting under uncertainty
  • Transformer-based fusion for complex cross-modal associations

OSIRIS IMPLEMENTATION

Python 3.12 + UV
Package management and virtual environments
PyTorch + Open3D
Deep learning components and 3D point cloud processing
Redis
Distributed session state management
FastAPI + gRPC + WebSocket
REST, streaming, and real-time data delivery

INTEGRATION POINTS

  • Provides human tracking to navigation and safety systems
  • Feeds presence data to fleet coordination modules
  • Outputs tracked positions for emergency response dashboards
08PAPERS

PROPRIETARY RESEARCH

  1. [01]Robot Flow Labs crown jewel IP — multi-modal human localization for GPS-denied environments