Lane‑Level Navigation Simulation: Road Scene Strategy Engine and Vehicle‑Behavior Modeling

1. Background In recent years, Amap’s automotive business has evolved from simple navigation to person‑vehicle co‑navigation, from a car‑mounted map to a car‑mounted map SDK, and finally to deep‑customized OEM projects. Despite diverse product forms, the goal of high‑quality in‑vehicle navigation remains unchanged.

Testing lane‑level navigation traditionally required extensive road testing, which became costly as OEM projects increased. To reduce road‑test costs, a simulation navigation feature was developed to emulate route planning and navigation on the device.

In 2021, Amap introduced lane‑level navigation with sub‑meter accuracy. The original simulation, based on standard map signals, offered 5‑10 m precision, insufficient for lane‑level scenarios that demand higher accuracy and complex test cases such as multi‑level overpasses and lane‑changing guidance. Therefore, a new simulation approach was needed to construct and reproduce lane‑level navigation scenes.

2. Problem Definition

2.1 Input

Lane‑level data provides high‑precision positioning signals required for lane‑level navigation.

Lane‑level signal replay records sensor data into a signal file that can be replayed on‑device to restore the entire lane‑level navigation process for offline analysis.

2.2 Output The goal is to build a simulation that can reconstruct realistic lane‑level signals from lane‑level data, enabling custom lane‑level navigation scene creation and meeting the following test requirements:

Cover the national high‑speed and city ordinary road lane networks with minimal route cost.

Allow custom construction of lane‑level navigation scenes.

Support custom simulation of lane‑level driving behavior.

3. Simulation Solution The solution consists of two engines:

Road Scene Strategy Engine : Determines which road segments a route must traverse and generates corresponding road segment data.

Vehicle Behavior Simulation Engine : Simulates vehicle actions such as lane changes and speed control based on surrounding information.

4. Road Scene Strategy Engine Road networks are split into small road segments with attributes (type, length, etc.). Graph algorithms are used for route planning. To cover large city road networks efficiently, a modified ant colony optimization (ACO) algorithm is applied. The algorithm evaluates each ant’s path using factors such as path length and “freshness” (unvisited nodes), adjusting pheromone weights accordingly. Early in the coverage process, longer routes are favored; later, routes that visit fresh nodes receive higher weight, achieving >94 % city road coverage with low route cost.

5. Vehicle Behavior Simulation Engine

5.1 Lane‑change strategy based on shortest path Each lane in a road segment is treated as a graph node. Links between lanes of consecutive segments are weighted by lane‑change cost, and Dijkstra’s algorithm finds the minimal‑weight lane‑change path. The cost considers whether lanes are aligned (cost = 1) or offset (cost = |j‑i|). The resulting lane‑change route is smoothed using cubic Bézier curves.

5.2 Dynamic speed simulation based on interval road shape Inspired by human driving behavior, the algorithm divides the trajectory into a reaction interval and a speed‑measurement interval . The reaction interval (≈2 s) models driver response time, while the speed‑measurement interval computes the maximum permissible speed based on road curvature, tire‑ground friction (μ≈0.8), and radius R using the formula v_max = √(μ·g·R). The algorithm continuously updates acceleration in the reaction interval, producing a realistic speed profile that handles acceleration, deceleration, and U‑turn scenarios.

By integrating the road scene strategy engine, lane‑change planning, and dynamic speed modeling, the system can generate high‑fidelity lane‑level navigation simulations that reduce testing cost, improve data quality verification, and support automated analysis for future quality improvements.