LiDAR Feature Learning for Robust Autonomous Vehicle Localization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing localization approaches for autonomous driving vehicles using LiDAR scanners require substantial engineering efforts and are scenario-dependent, lacking effectiveness in 3D geometry-related tasks such as localization.
Innovation Solution
A learning-based LiDAR localization system employing convolutional neural networks (CNNs) and recurrent neural networks (RNNs) to extract point cloud features, construct a cost volume, and estimate the optimal pose of the vehicle by matching online LiDAR points with a pre-built point cloud map, reducing the need for handcrafted modules and human labeling efforts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing localization approaches using LiDAR scanners are employed, then localization accuracy can be achieved, but substantial engineering efforts and handcrafted modules are required
Solution Approach 1:
The patent replaces traditional handcrafted geometric algorithms and manual feature engineering with deep learning-based neural networks. The system uses trained neural network models to automatically extract features from LiDAR point clouds and perform localization, substituting the mechanical process of manual algorithm development and tuning with an intelligent system that learns optimal features from data.
2Reliability
If existing localization approaches are used, then robustness across scenarios can be achieved, but strong preferences over certain scenarios and handcrafted modules are needed
Solution Approach 1:
The patent creates a universal localization system using deep learning models that can adapt to multiple driving scenarios without requiring scenario-specific handcrafted modules. The neural networks are trained on diverse data to learn generalizable features that work across different environments, road types, and conditions, making the system versatile and adaptable rather than scenario-dependent.
Solution Approach 2:
The system changes the approach from fixed handcrafted parameters to learned parameters through neural network training. The model learns optimal feature extraction parameters and localization strategies from training data, allowing it to adapt to different scenarios by leveraging patterns learned during training rather than relying on pre-defined scenario-specific parameters.
3Loss of information
If neural networks are used for semantics processing, then good semantic results are achieved, but they lack effectiveness in solving 3D geometry tasks like localization
Solution Approach 1:
The patent merges semantic processing capabilities with 3D geometry processing in a unified deep learning framework. The system uses neural networks that simultaneously handle both semantic understanding of the environment and precise 3D localization tasks, combining the strengths of semantic AI with geometric accuracy in an integrated architecture rather than treating them as separate problems.
Data Source
AI summary
In one embodiment, a method for extracting point cloud features for use in localizing an autonomous driving vehicle (ADV) includes selecting a first set of keypoints from an online point cloud, the online point cloud generated by a LiDAR device on the ADV for a predicted pose of the ADV; and extracting a first set of feature descriptors from the first set of keypoints using a feature learning neural network running on the ADV, The method further includes locating a second set of keypoints on a pre-built point cloud map, each keypoint of the second set of keypoints corresponding to a keypoint of the first set of keypoint; extracting a second set of feature descriptors from the pre-built point cloud map; and estimating a position and orientation of the ADV based on the first set of feature descriptors, the second set of feature descriptors, and a predicted pose of the ADV.