Humanoid Robotics (ISO 13849)
OmniSensor Integration: Multi-Sensor Fusion Topology
Software-in-the-Loop (SIL) test system with LiDAR (60%) + Vision (40%) fusion, fallback mechanisms on hardware failure, <100ms latency under occlusion
The Hidden Problem: Sensor Weights Hidden in Embedded C++
- Requirement: "Reliable human detection <100ms, ISO 13849 compliant"
- Implementation: LiDAR 60% weight + Vision 40% weight (Fusion)
- The Problem: Weights are hardcoded in HAL code, not traceable
- Safety Question: "If the ML backend crashes — can the requirement still be met?"
- Manual Answer: Safety engineer must review C++ + ROS2 architecture (4+ hours)
- Graph Answer: Instant: "LiDAR fallback active, +40ms latency but <100ms" ✓
The Vimpact Graph: Automatic Dependency Analysis
- All Sensor Modules as Nodes: ML_INFERENCE_ENGINE, CPU_FALLBACK, LIDAR_DETECTOR, VISION_DETECTOR
- Edges with Weights & Fallback: weight: 0.4, has_fallback: true, latency_impact: "+40ms"
- Latency Budget as Graph Constraint: REQ_100MS depends_on [ML_INFER + FUSION + SAFETY_LOGIC]
- Impact Analysis: "Remove ML_BACKEND" → Graph computes paths → 60ms (OK) or 120ms (FAIL)
- Compliance Verification: Does every fallback path have an ISO 13849 test? Graph validates!
Real Sensor-Fusion Topology
SAFETY REQUIREMENT (Graph-Knoten): "Human detection <100ms, confidence >95%, under occlusion" │ verified_by (OpenSCENARIO) ▼ TEST SCENARIO: "Person behind obstacle (60% occlusion) + twilight" │ depends_on (Weighted Sensor Fusion) ├─▶ 3D-Sensor (60% Gewicht) — 16-Ring, 180° FOV, 10Hz ├─▶ RGB-Depth Kamera (40% Gewicht) — 120° HFOV, RealSense └─▶ Kalman-Filter Tracking — 5-Frame Memory, Velocity-Estimation
Graph Node: ML Inference Engine
{
"node_id": "HAL_ML_INFERENCE",
"type": "inference_engine",
"framework": "Quantized Vision Transformer (local)",
"edges": [
{ "to": "SENSOR_FUSION_NODE", "type": "powers", "weight": 0.4 },
{ "to": "HAL_CPU_FALLBACK", "type": "has_fallback", "latency_penalty_ms": 100 }
],
"properties": { "latency_ms": 15, "failure_rate": "< 1e-8/h" }
}