Geometry-Aware Polyline Map Generation for Smooth Road Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current map generation models for autonomous driving assistance systems (ADAS) often produce maps with irregular-shaped pedestrian crossings and randomly-shaped center lines and road boundaries that do not accurately represent the actual features in the environment, leading to navigation challenges for autonomous vehicles.
Innovation Solution
A map generation model is trained using a loss function that includes polygon and polyline losses to promote convexity and smoothness, ensuring that features are represented as convex polygons and smooth polylines, thereby improving map accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional deep-learning methods are used to train map generation models, then the models can generate maps from sensor data, but the generated maps contain irregular-shaped pedestrian crossings and randomly-shaped center lines and road boundaries that do not accurately represent actual features
Solution Approach 1:
The patent changes the training parameters by introducing geometry-aware loss functions that specifically target shape accuracy. The loss function includes terms that penalize deviations from expected geometric properties (convexity for polygons, smoothness for polylines), transforming the training objective from generic pixel-level accuracy to geometry-aware accuracy that ensures reliable feature representation.
Solution Approach 2:
The patent implements feedback mechanisms through the loss function that continuously evaluates the geometric properties of generated map features and adjusts the model parameters accordingly. The geometry-aware loss provides feedback on shape deviations, guiding the model to produce more accurate representations of pedestrian crossings, center lines, and road boundaries.
2Adaptability or versatility
If the map generation model produces irregular-shaped features, then it can capture complex environmental variations, but the maps become less useful for ADAS navigation and safety applications
Solution Approach 1:
The patent applies local quality by imposing specific geometric constraints on different types of map features. Pedestrian crossings are constrained to be convex polygons, while center lines and road boundaries are constrained to be smooth polylines. This local geometric quality assurance maintains adaptability to environmental variations while ensuring reliability for navigation applications.
Solution Approach 2:
The patent uses curvature constraints to ensure smooth polylines for center lines and road boundaries. By penalizing abrupt changes in direction and enforcing smooth transitions, the model generates reliable geometric representations that are suitable for navigation while still capturing the necessary environmental variations.
Data Source
AI summary
Certain aspects of the present disclosure provide a method for generating a map. A method generally includes obtaining sensor data from one or more sensors, the sensor data corresponding to an environment comprising one or more features including a first feature having one or more changes in curvature; inputting the sensor data to a map generation model; and obtaining as output from the map generation model a vectorized map of the one or more features of the environment, wherein the vectorized map represents the first feature as a polyline comprising a plurality of line segments, wherein a comparison of changes of slope between consecutive line segments of the polyline to slopes of the plurality of line segments of the polyline satisfies a threshold.