Lane-Level Map Matching Using Hidden Markov Lane States
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current map matching solutions for autonomous vehicles are not accurate enough, particularly due to noise in GPS signals and imperfect sensor data, and do not provide robust lane-level localization.
Innovation Solution
A method for lane-level map matching that uses a state space model, specifically a Hidden Markov Model, to determine the most probable path of a vehicle by combining vehicle data, sensor data, and map data, which includes forming a set of states representing candidate lanes and defining costs for transitions between these states based on predefined probabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional GPS-based map matching is used, then the system is simple to implement, but the measurement precision is insufficient due to noise in GPS signals
Solution Approach 1:
The patent combines multiple data sources (GPS coordinates, sensor observations of road references, and map data) into a unified map matching system. By merging these complementary data sources, the system achieves lane-level localization accuracy that exceeds what any single source could provide alone, while the integrated approach manages complexity through coordinated processing.
Solution Approach 2:
The patent introduces a probabilistic model as an intermediary layer between raw GPS/sensor data and final lane identification. This probabilistic framework mediates the integration of uncertain measurements with map data, enabling accurate lane-level localization by computing the probability of the vehicle being in each candidate lane rather than directly mapping noisy inputs to discrete lanes.
2Reliability
If sensor data fusion is implemented to improve robustness, then the reliability improves, but the computational complexity increases
Solution Approach 1:
The patent computes probabilities for multiple candidate lanes simultaneously rather than sequentially eliminating options. By maintaining and evaluating probabilistic estimates for all plausible lanes at each time step, the system achieves robustness to measurement errors and data inconsistencies, accepting the computational cost of evaluating multiple hypotheses in parallel.
Solution Approach 2:
The system uses probabilistic feedback from sensor observations to continuously update lane probability estimates. Each new sensor measurement feeds back into the probabilistic model, which adjusts the likelihood of each candidate lane based on how well the observations match expected patterns for that lane, thereby improving reliability through iterative refinement.
3Measurement precision
If lane-level precision is achieved through multiple data sources, then the measurement precision improves, but the difficulty of detecting and measuring increases
Solution Approach 1:
The patent transforms the map matching problem from directly determining vehicle position to computing probabilities across discrete lane states. By changing the parameter representation from continuous coordinates to discrete probabilistic lane assignments, the system achieves lane-level precision while managing the complexity of integrating multiple data sources through a standardized probabilistic framework.
Data Source
AI summary
The present disclosure relates to lane-level map matching for a vehicle. The method includes receiving vehicle data including a geographical position of the vehicle, a heading of the vehicle, and a speed of the vehicle and receiving sensor data from a perception system of the vehicle. The sensor data includes information about a position of at least one road reference in a surrounding environment of the vehicle. Operations include receiving map data including a lane geometry of the surrounding environment of the vehicle, the lane geometry including a set of candidate lanes. Operations include forming a state space model including a set of states. Each state of the set of states represents a candidate lane of the set of candidate lanes, and defining a cost for going from each state to every other state of the set of states based on the received vehicle data and the received sensor data.