How High‑Precision Positioning Enables Lane‑Level Navigation: Techniques and Roadmap

On October 30, Huawei, Amap (Gaode), and Qianxun launched the industry’s first lane‑level navigation app for smartphones, highlighting the maturity of high‑precision positioning technologies. The article uses Gaode’s work in lane‑level navigation and autonomous driving to discuss the evolution of positioning techniques and practical implementations.

1. Overview of Gaode’s Positioning Technology

Gaode’s core positioning stack combines GNSS, inertial sensors, Wi‑Fi/Bluetooth, and cellular signals to provide basic location coordinates. For navigation, map‑matching is applied to determine the vehicle’s lane and heading, with specialized models handling complex scenarios such as elevated roads and auxiliary lanes.

The complete system follows a "cloud + edge + data" architecture, supported by a quality‑iteration platform that continuously updates each module.

2. Path to High‑Precision Positioning

To achieve lane‑level navigation (sub‑meter accuracy) and autonomous driving (centimeter‑level accuracy), positioning technologies are categorized into three groups:

Dead‑reckoning (DR) : Propagates position from a known start using motion direction and distance. Accuracy degrades over time, depending on sensor quality and model compensation.

Geometric positioning : Measures distances or angles to known reference points (RSS, TOA, AOA, TDOA). Accuracy is limited by measurement errors, signal range, and the number/distribution of reference nodes.

Feature‑based positioning : Matches environmental features (Wi‑Fi fingerprints, magnetic fields, images, LiDAR point clouds) to a pre‑built map or uses SLAM for relative positioning. Accuracy depends on feature density, quality, and distinctiveness.

A high‑precision solution typically combines at least one absolute technique (e.g., GNSS RTK or LiDAR matching) with relative methods (DR, visual SLAM) to compensate for environmental constraints.

3. Business Scenarios and Requirements

Different navigation use cases demand varying precision and reliability:

Standard turn‑by‑turn navigation : ~10 m accuracy.

Lane‑level navigation : < 1 m accuracy to identify the current lane.

Advanced driver assistance / autonomous driving : < 0.2 m lateral accuracy, high reliability, and real‑time integrity monitoring.

Additional requirements include reliability (error detection), availability (accurate confidence radius), computational load, and stability across diverse environments.

4. Lightweight Integrated Fusion Solution

The proposed engine fuses RTK‑GNSS, visual semantic matching, and IMU/vehicle‑model DR. Visual semantic matching uses camera‑detected lane markings and ground signs to align with high‑definition maps, keeping computational cost low. The overall pipeline maintains road‑level and lane‑level outputs while supporting partial sensor availability.

Key algorithmic enhancements for the particle‑filter‑based fusion include:

Hypothesis‑driven particle dimension reduction to lower computation.

Hierarchical normalization to mitigate particle degeneration caused by minor system errors.

Context‑aware posterior confidence estimation to handle missing or erroneous sensor confidence.

Signal‑window based handling of latency and out‑of‑order measurements.

Use of high‑precision satellite positioning and HD maps for sensor calibration, improving DR performance.

This solution has been deployed on an L3‑level autonomous vehicle for large‑scale testing.

5. Tightly‑Coupled Multi‑Sensor SLAM for Complex Scenes

In environments where GNSS is unavailable (indoor, parking lots, dense urban intersections), a tightly‑coupled SLAM framework combines low‑cost sensors (GNSS, IMU, cameras) into an optimized model that jointly estimates pose and sensor biases.

Evaluations on the EuRoc and Kaist datasets show more than a two‑fold improvement in positional accuracy compared with standard visual‑IMU or visual‑IMU‑GNSS fusion approaches. Future work focuses on optimizing computational load for mobile and in‑vehicle platforms.

6. Conclusions and Outlook

High‑precision positioning has evolved from specialized surveying tools to consumer‑grade solutions that can support lane‑level navigation and emerging autonomous driving functions. Future directions include lower‑cost PPP‑RTK, 5G‑assisted positioning, affordable LiDAR integration, and deeper multi‑sensor fusion to achieve full‑scene coverage across indoor, outdoor, vehicular, and pedestrian contexts.