Introduction
Self-driving cars need to know where they are, precisely, continuously, and reliably. GNSS is a critical part of the solution, but it's not enough on its own.
The Autonomous Vehicle Challenge
Requirements: lane-level accuracy (10–30 cm), continuous availability (no gaps), immediate initialization, integrity (must know when it's wrong), and security (can't be spoofed). Environment: urban canyons, tunnels, tree cover, and multipath everywhere. No single sensor meets all these requirements.
The Sensor Fusion Solution
| Sensor | Strengths | Weaknesses |
|---|---|---|
| GNSS | Absolute position, long-term stable | Blocked by buildings, multipath |
| INS | Continuous, short-term accurate | Drifts over time |
| LiDAR | 3D map, obstacle detection | Weather sensitive, expensive |
| Camera | Lane markings, signs | Lighting dependent |
| Radar | Velocity, all-weather | Low resolution |
| Wheel odometry | Speed, distance | Slip, tire wear |
A Kalman filter combines all of these with GNSS/INS for the optimal position estimate at all times.
What GNSS Provides, and Where It Struggles
GNSS provides global absolute positioning, lane-level accuracy with RTK/PPP, time synchronization, and heading with dual-antenna setups. It struggles in urban canyons (blocked signals), with multipath (reflections), and with instant initialization (RTK takes seconds to converge).
Automotive-Grade GNSS
Requirements: multi-frequency (L1+L5), multi-constellation (GPS+Galileo+BeiDou), RTK/PPP capability, INS integration, functional safety (ISO 26262), and automotive qualification (AEC-Q100). Common modules include u-blox F9 series and Septentrio for high-end applications.
Map Matching and HD Maps
High-Definition maps provide a centimetre-accurate road model with lane markings, curbs, signs, slope, and curvature. Map matching compares the GNSS position to the map to constrain the vehicle to a plausible location, determine the correct lane, and provide an integrity check, if the map and GNSS disagree, something is wrong.
Dead Reckoning During GNSS Outages
In tunnels and urban canyons, the vehicle uses wheel speed sensors (ABS), gyroscope (yaw rate), accelerometer, and reverse camera (visual odometry) to navigate. Performance: good INS achieves <1% drift (1 m error after 100 m); standard automotive: 2–5% drift. This enables navigation through several minutes of tunnel.
Integrity: Knowing When It's Wrong
Autonomous vehicles must know when GNSS is unreliable. Integrity monitoring uses RAIM-like algorithms and consistency checks across all sensors to produce protection levels (95% confidence bounds). On integrity failure, the system degrades gracefully, slowing down, pulling over, or performing a safe stop.
Current State and Future
- Today: Highway autopilot with lane keeping and adaptive cruise; requires driver supervision; GNSS + map + cameras
- Near future (2025–2028): Urban autonomous driving with improved sensor fusion
- Long term (2030+): Full autonomy anywhere with V2X communication augmentation and possibly LEO GNSS
Vital Points
- Autonomous vehicles need continuous, reliable positioning
- GNSS provides absolute position but needs help in cities
- Sensor fusion combines GNSS, INS, cameras, LiDAR, and radar
- HD maps and map matching provide lane-level context
- Dead reckoning bridges GNSS outages
- Integrity monitoring is essential for safety
- No single sensor is sufficient, redundancy is key