Autonomous Drivable Surface Estimation via Lateral Slice Grouping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional autonomous driving systems struggle to identify drivable surfaces outside marked lane boundaries, such as in sections without lane markings, during construction, or in complex road scenarios, leading to inaccurate trajectory execution.
Innovation Solution
The system uses probe data, satellite imagery, and road/navigation graphs to discretize roadways into lateral slices, determine features, group slices into lane groups, and classify each slice based on these groups, employing a CNN network for accurate lane boundary estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional autonomous driving systems rely only on marked lane boundaries to identify drivable surfaces, then the system complexity remains low, but the measurement precision of drivable surface identification deteriorates in sections without lane markings or during construction
Solution Approach 1:
The patent divides the roadway into multiple lateral slices and further segments each slice into different surface types (marked lane, unmarked lane, shoulder, etc.). This segmentation allows the system to analyze and identify drivable surfaces in each segment independently, improving overall identification accuracy without requiring a single complex model to handle all scenarios.
Solution Approach 2:
The patent transitions from relying solely on two-dimensional lane markings to incorporating multi-dimensional data including satellite imagery, probe data from multiple vehicles, and road navigation graphs. By adding these additional dimensions of data, the system achieves better drivable surface identification accuracy especially in areas without clear lane markings.
2Reliability
If the system uses only marked lane boundaries for trajectory execution, then the ease of operation is high, but the reliability of trajectory execution deteriorates in complex road scenarios such as construction zones or unmarked sections
Solution Approach 1:
The patent creates a universal drivable surface identification system that can handle multiple road scenarios (marked lanes, unmarked lanes, construction zones, shoulders) using a single integrated approach. The system processes probe data, satellite imagery, and road graphs through a unified methodology, making it reliably applicable across diverse road conditions without requiring separate specialized systems for each scenario.
Solution Approach 2:
The system incorporates feedback mechanisms by using probe data from multiple vehicles and iteratively refining the drivable surface identification. The histogram-based approach aggregates information from multiple sources and uses confidence thresholds to validate identifications, creating a feedback loop that improves reliability while maintaining operational simplicity through automated decision-making.
3Manufacturing precision
If the system discretizes probe data into multiple lateral slices and groups them into lane groups, then the manufacturing precision of drivable surface mapping is improved, but the loss of time for data processing increases
Solution Approach 1:
The patent segments the continuous probe data into discrete lateral slices of fixed width (e.g., 0.5 meters), which allows parallel processing of each slice independently. This segmentation enables the system to achieve precise drivable surface mapping through systematic analysis of each segment while facilitating efficient computation through divide-and-conquer methodology.
Solution Approach 2:
The system performs preliminary actions by pre-defining slice widths, confidence thresholds, and grouping criteria before actual data processing. The histogram generation and lane group identification are prepared in advance with predetermined parameters, which reduces computational complexity during real-time processing and minimizes data processing time while maintaining mapping precision.
Data Source
AI summary
Systems and methods are provided for determining lane groups for use in autonomous driving. The system can receive probe data of an autonomous vehicle traveling on a roadway and discretize the probe data into a plurality of lateral slices of the roadway. Features can be determined, the features being associated with the plurality of lateral slices. A portion of the plurality of lateral slices can be grouped into a lane group based on the features. Each slice can be classified based on the lane group.