PPP-RTK Positioning for Autonomous Driving

Problem

L2+ autonomous driving requires lane-level positioning (< 0.3m), but mass-production vehicles face a fundamental set of contradictions.

Technical Contradictions

• Traditional RTK achieves the required accuracy but depends on dense base station networks — per-vehicle service cost is too high for millions of units
• Pure vision/IMU solutions are cheap but have systematic blind spots: tunnel exits, under overpasses, featureless highways
• HD map solutions degrade over time, and regulatory pressure continues to tighten

Product Contradictions

• OEMs demand nationwide coverage, not regional patches
• Consumers have zero tolerance for positioning failures at highway speed
• Tier-1 supply chain requires module cost under ¥200

Market Timing

• L2+ penetration: 15% → 40% (2023-2025), positioning becomes a hard dependency
• BDS-3 global constellation complete, PPP-RTK moves from theory to commercial viability

---

Solution

Product Definition: Cloud-Edge Collaborative Positioning Service

Not selling hardware — defining a positioning-as-a-service product:

Cloud (Service Layer):

• National CORS network ingestion
• PPP-RTK engine generating SSR corrections
• Integrity monitoring with protection levels

Vehicle (Terminal Layer):

• GNSS module + IMU tight coupling
• Integrity decision engine
• Wheel speed + visual fusion fallback

Key Product Decisions

Decision 1: SSR Architecture vs OSR Architecture

Dimension	OSR (Traditional RTK)	SSR (PPP-RTK)
Bandwidth	~10 KB/s	< 1 KB/s
Base Station Dependency	Strong (< 30km)	Weak (regional broadcast)
Per-Vehicle Cost	High (point-to-point)	Low (broadcast model)
Scalability	Poor (linear growth)	Excellent (marginal cost → 0)

→ Chose SSR. The core product requirement is scaling to millions, not peak precision.

Decision 2: Integrity Over Accuracy

The industry default was to chase higher accuracy. Product analysis revealed:

• Autonomous driving doesn’t need “always 2cm” — it needs to know when it’s unreliable
• Defined Protection Level as the primary output metric
• When PL > threshold, proactively signal the fusion system to downgrade to vision/IMU

This decision directly improved pass rates in OEM safety reviews.

Decision 3: Single-Frequency + Cloud Compensation

• Dual-frequency module: accurate but ¥150+
• Single-frequency module: ¥50, but higher multipath error
• Final approach: single-frequency + cloud-side multipath model, converting hardware cost to algorithm cost
• Cloud-side models iterate continuously; terminals upgrade via OTA

Product Tiering

• Tier 1 — Basic (sub-meter): L2 lane keeping. ¥3/vehicle/month
• Tier 2 — Precision (decimeter + integrity): L2+ highway/urban NOA. ¥8/vehicle/month
• Tier 3 — Safety-critical (integrity + functional safety cert): L3+. Custom pricing

---

Impact

Commercial:

• Integrated into XX+ mass-production vehicle models
• 70% cost reduction vs traditional RTK approach
• Cloud service gross margin > 65%

Technical:

• Nationwide convergence time < 30s (industry average 60s+)
• Availability > 99.5% including urban canyon scenarios
• First PPP-RTK service to achieve ASIL-B certification

Industry:

• Shifted OEM mindset from “buy a module” to “subscribe to a positioning service”
• Architecture adopted as reference design by multiple Tier-1 suppliers

---

Reflection

1. Scale thinking > peak performance: Choosing SSR over OSR isn’t technically optimal — it’s product optimal. Technology selection should be driven by deployment constraints, not lab benchmarks.

2. Safety narrative > accuracy narrative: OEMs don’t fear inaccuracy — they fear not knowing it’s inaccurate. Integrity reframed the entire value proposition.

3. Convert hardware cost to software cost: Single-frequency + cloud compensation trades non-iterable hardware for iterable algorithms — giving the product continuous evolution capability.