HD Map Generation from Aligned Geospatial Observations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional methods for 3D road geometry modeling and feature detection in autonomous vehicles are resource-intensive and time-consuming, often requiring manual or semi-automated analysis of large data sets, leading to inaccurate map reconstruction and safety concerns due to unreliable feature detection.
Innovation Solution
A system using a processor and memory configuration to align and process geospatial observations through iterative attention models with Gated Recurrent Units and Set Transformers, generating high-definition maps for autonomous vehicle navigation, incorporating point and linear objects like signs and road boundaries.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional manual or semi-automated methods are used for 3D road geometry modeling and feature detection, then measurement precision may be maintained through human analysis, but productivity is severely reduced due to time-consuming processing and resource intensity
Solution Approach 1:
The patent replaces manual mechanical analysis methods with automated machine learning systems. Specifically, it employs neural networks and computer vision algorithms to automatically detect features, extract geometries, and generate maps from sensor data, eliminating the need for human measurement and calculation while maintaining high precision through learned patterns from training data
Solution Approach 2:
The system implements self-service through automated processing pipelines where the machine learning models independently perform feature detection, geometry extraction, and map generation without human intervention. The system processes sensor data from multiple sources, aligns trajectories, and produces HD maps autonomously, enabling continuous operation at high speed
2Reliability
If feature detection systems operate with high confidence thresholds, then reliability of detected features improves, but productivity decreases due to missed detections requiring re-scan
Solution Approach 1:
The patent merges data from multiple sensor sources including LiDAR, cameras, and radar to create complementary views of the environment. By combining these diverse sensor inputs, the system achieves both high reliability through cross-validation of detected features and high productivity through parallel processing of multiple data streams simultaneously
Solution Approach 2:
The system implements feedback mechanisms where detected features and generated maps are continuously validated against new sensor observations. Discrepancies trigger targeted re-detection or correction, allowing the system to maintain high reliability while avoiding unnecessary full re-scans, thus preserving productivity
3Measurement precision
If multiple sensor trajectories are processed independently to maintain data integrity, then measurement precision is preserved, but device complexity increases due to separate processing pipelines
Solution Approach 1:
The patent merges multiple independent trajectory processing pipelines into a unified map generation system. Individual sensor trajectories are processed to extract features and geometries, then these results are integrated into a single consistent HD map representation, reducing overall system complexity while preserving the precision benefits of independent processing through standardized integration protocols
Data Source
AI summary
A method, apparatus and computer program product are provided for learning to generate maps from raw geospatial observations from sensors traveling within an environment. Methods may include: receiving a plurality of sequences of geospatial observations from discrete trajectories; aligning the discrete trajectories generating aligned geospatial observations; concatenating the aligned geospatial observations; performing attentional clustering on the concatenated, aligned geospatial observations to obtain a set of entities with feature dimensionality; processing the set of entities through an iterative attention model incorporating a Gated Recurrent Unit gating pattern to obtain attentional layer outputs; generating, from one or more Set Transformers, a feature set of map object geometries based, at least in part, on the attentional layer outputs; updating a map geometry based on the feature set from the Set Transformers generating an updated map geometry; and provide for navigational assistance or at least semi-autonomous vehicle control based on the updated map geometry.