Physics-Informed Human Understanding

Unlike traditional pose estimation that relies on statistical pattern matching, AuraSense uses physics and biomechanics to truly understand human movement.

Traditional Approach

  • 2D/3D keypoints from neural networks
  • Black-box predictions with no physical validity
  • Jitter and physically impossible poses
  • Limited to what the model was trained on

AuraSense Approach

  • Physics simulation with mass, inertia, joint constraints
  • Inverse kinematics ensures biomechanical validity
  • Smooth, physically plausible motion
  • Explainable metrics (velocity, joint angles, tremor frequency)

System Architecture

┌─────────────────────────────────────────────────────────────────────────────┐
│                           INPUT LAYER                                       │
├─────────────────────────────────────────────────────────────────────────────┤
│  Video Stream (webcam/depth) │ Audio Stream │ Biosensors (optional)        │
└─────────────────────────────────────────────────────────────────────────────┘
                                    │
                                    ▼
┌─────────────────────────────────────────────────────────────────────────────┐
│                       PROCESSING LAYER                                      │
├─────────────────────────────────────────────────────────────────────────────┤
│  ┌──────────────────┐  ┌──────────────────┐  ┌──────────────────┐          │
│  │ Initial Pose     │  │ Physics Engine   │  │ Biomechanics     │          │
│  │ Detection        │─▶│ (Inverse IK)     │─▶│ Validation       │          │
│  │ (Neural Net)     │  │                  │  │                  │          │
│  └──────────────────┘  └──────────────────┘  └──────────────────┘          │
│                                                                             │
│  ┌──────────────────┐  ┌──────────────────┐  ┌──────────────────┐          │
│  │ Audio Analysis   │  │ Facial Sentiment │  │ Multi-Modal      │          │
│  │ (MFCC, VAD)      │  │ (Valence/Arousal)│  │ Fusion           │          │
│  └──────────────────┘  └──────────────────┘  └──────────────────┘          │
└─────────────────────────────────────────────────────────────────────────────┘
                                    │
                                    ▼
┌─────────────────────────────────────────────────────────────────────────────┐
│                         OUTPUT LAYER                                        │
├─────────────────────────────────────────────────────────────────────────────┤
│  Raw Kinematics │ Digital Biomarkers │ Real-time Feedback │ Session Export │
└─────────────────────────────────────────────────────────────────────────────┘
          
<100ms
End-to-end latency
60fps
Real-time processing
33+
Skeletal joints tracked
Browser
No install required

Built With

🦀

Rust

Core engine, maximum performance

WebTransport

QUIC-based low-latency streaming

🧠

WebGPU

Browser-native GPU acceleration