Intersection Pose Detection From Live Sensors Without HD Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems for autonomous vehicles rely on high-definition maps for intersection detection, requiring extensive manual labeling and decreasing scalability, especially in complex urban environments.
Innovation Solution
Utilizing live perception from vehicle sensors, such as cameras and LIDAR, to detect intersection poses and generate paths through machine learning algorithms, allowing real-time or near real-time navigation without prior knowledge of the intersection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If map-based solutions with manual labeling are used for intersection detection, then intersection structure information can be obtained, but the complexity of labeling increases and scalability decreases
Solution Approach 1:
The patent replaces the manual mechanical labeling process with an automated machine learning system. The neural network automatically detects and labels intersection structures from sensor data, eliminating the need for manual annotation of HD maps while maintaining detection accuracy.
Solution Approach 2:
The system enables the vehicle to autonomously detect and understand intersection structures using its own sensors and onboard processing capabilities, without relying on pre-labeled external maps. The vehicle serves itself by generating its own environmental understanding in real-time.
2Loss of information
If HD maps with detailed annotations are used, then comprehensive intersection information is available, but the system becomes less scalable to new regions
Solution Approach 1:
The system transitions from static pre-labeled maps to dynamic real-time detection. The neural network continuously processes sensor data to adapt to new intersections and regions on-the-fly, enabling the system to scale to any location without requiring prior manual labeling efforts.
Solution Approach 2:
The system performs preliminary detection and understanding of intersection structures using sensor data and machine learning before navigation decisions are required, eliminating the need for pre-existing annotated maps while maintaining complete intersection information.
3Measurement precision
If conventional map-based approaches are used, then intersection poses can be determined, but computational complexity and processing time increase
Solution Approach 1:
The patent replaces computationally intensive map matching and interpolation operations with a streamlined neural network inference process. The trained model directly predicts intersection poses from sensor inputs, significantly reducing processing time while maintaining accuracy.
Solution Approach 2:
The system changes the computational parameters by using pre-trained neural network models that have already learned intersection patterns during training. During deployment, the system performs efficient inference with fixed parameters rather than executing complex real-time map processing algorithms.
Data Source
AI summary
In various examples, live perception from sensors of a vehicle may be leveraged to generate potential paths for the vehicle to navigate an intersection in real-time or near real-time. For example, a deep neural network (DNN) may be trained to compute various outputs—such as heat maps corresponding to key points associated with the intersection, vector fields corresponding to directionality, heading, and offsets with respect to lanes, intensity maps corresponding to widths of lanes, and/or classifications corresponding to line segments of the intersection. The outputs may be decoded and/or otherwise post-processed to reconstruct an intersection—or key points corresponding thereto—and to determine proposed or potential paths for navigating the vehicle through the intersection.