Probabilistic Online Mapping for Autonomous Vehicles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional online mapping methods for autonomous agents face challenges in creating and maintaining high-definition maps, which are difficult and expensive to update, and suffer from sparse feature-based methods and uncertainties in learning-based methods, especially in crowded traffic and poor weather conditions.
Innovation Solution
A learning-based online mapping system that uses a deep neural network to generate a probabilistic map of surroundings by localizing an ego vehicle relative to an offline feature map, querying surrounding features, and updating the feature map periodically to improve the accuracy and reliability of map information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If conventional feature-based mapping methods are used, then map creation is simpler, but map information becomes sparse and inaccurate in complex environments
Solution Approach 1:
The system transforms the mapping approach by changing from binary feature detection to probabilistic parameter estimation. Instead of detecting discrete features, the system estimates spatial parameters (position, orientation, semantics) with associated probability distributions, enabling accurate representation even when features are occluded or ambiguous.
Solution Approach 2:
The patent introduces an intermediary probabilistic representation layer between sensor data and map information. This probabilistic map serves as a mediator that captures uncertainty and ambiguity, allowing the system to handle complex environments without requiring precise feature detection.
2Measurement precision
If learning-based mapping methods are used, then map information becomes richer and more accurate, but uncertainties increase especially in crowded traffic and poor weather conditions
Solution Approach 1:
The system implements feedback mechanisms where the probabilistic map continuously refines itself based on sensor observations. When uncertainties are detected (e.g., in crowded traffic or poor weather), the system queries for additional information and updates the probabilistic representations, gradually improving reliability without sacrificing the richness of map information.
3Measurement precision
If annotated high definition maps are created, then surrounding vehicle information becomes more accurate, but map creation and maintenance become difficult and expensive
Solution Approach 1:
The system enables self-service mapping where the probabilistic map automatically updates itself using data from the vehicle's sensors and communications with other vehicles. Instead of requiring manual annotation and maintenance, the map continuously refines itself through collaborative filtering and sensor fusion, significantly reducing the complexity and cost of map creation and maintenance.
Solution Approach 2:
The patent uses copying by aggregating probabilistic map data from multiple vehicles and sensors to create a comprehensive surrounding environment representation. This collaborative copying approach allows the system to achieve HD map accuracy without requiring manual annotation, as the map is constructed by synthesizing information from multiple sources.
4Device complexity
If feature-based mapping is used, then system complexity is lower, but handling ambiguities in road structures becomes difficult
Solution Approach 1:
The system changes from detecting discrete features to estimating continuous spatial parameters with probability distributions. This parameter transformation enables the system to handle ambiguities in road structures (e.g., lane configurations, intersection geometries) by representing them as probabilistic ranges rather than fixed features, increasing adaptability without significantly increasing system complexity.
Data Source
AI summary
A method for an online mapping system includes localizing a location of an ego vehicle relative to an offline feature map. The method also includes querying surrounding features of the ego vehicle based on the offline feature map. The method further includes generating a probabilistic map regarding the surrounding features of the ego vehicle queried from the offline feature map.