Autonomous Vehicle Localization Using Map-Sensor Pose Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current localization systems for autonomous vehicles are either expensive and computationally demanding for high accuracy or sacrifice accuracy for cost and power efficiency, failing to provide timely and reliable location information, especially in complex environments.
Innovation Solution
A method and system that processes both offline map data and real-time sensor data to isolate relevant information for continuous localization, using a multi-processor approach to filter and match data, correct pose estimation, and predict future vehicle characteristics, incorporating LIDAR, GPS, and optical data for accurate and efficient localization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If expensive sensor suites are used for high accuracy localization, then localization precision is improved, but device cost and computational power requirements increase
Solution Approach 1:
The patent segments the localization task into multiple processing stages: data collection from multiple sensors, offline map creation, real-time sensor data processing, feature matching, and pose estimation. This segmentation allows each stage to be optimized independently, achieving high accuracy without requiring all sensors to operate at maximum capacity simultaneously, thus reducing overall system complexity and cost.
Solution Approach 2:
The patent performs preliminary actions by creating offline maps with pre-processed environmental features before the vehicle needs localization. These pre-created maps contain organized spatial information that can be quickly matched against real-time sensor data, eliminating the need for complex real-time processing of all raw sensor information and reducing computational power requirements during critical localization moments.
2Measurement precision
If expensive sensor suites are used for high accuracy localization, then localization precision is improved, but power consumption increases
Solution Approach 1:
The patent implements periodic action by updating localization estimates at discrete time intervals rather than continuously processing all sensor data at maximum rate. The system collects sensor data continuously but performs computationally intensive matching and pose estimation operations periodically, allowing power-intensive operations to be concentrated in controlled bursts rather than sustained at maximum level, thus reducing overall power consumption while maintaining accuracy.
3Ease of operation
If GPS-based localization is used, then basic navigation information is provided, but accuracy is insufficient for precise vehicle location requirements
Solution Approach 1:
The patent merges GPS data with data from multiple other sensors (lidar, cameras, IMU) and combines this with offline map information to create a multi-source localization system. The GPS provides coarse location information that helps select relevant offline map sections, while the other sensors provide fine-grained features for precise matching, achieving both ease of operation and high precision simultaneously.
4Measurement precision
If map-based localization with high resolution image-based maps is used, then localization accuracy is improved, but storage requirements increase
Solution Approach 1:
The patent extracts only the essential localization features from high-resolution maps during offline processing, storing compressed representations rather than complete high-resolution images. The system extracts spatial relationships, landmark positions, and geometric features that are sufficient for accurate localization while removing redundant visual detail, thus maintaining localization accuracy while dramatically reducing storage requirements.
Data Source
AI summary
A system and method for localizing an autonomous vehicle using mapped and real-time data. Mapped data and real-time data are scan matched. Characteristics of the autonomous vehicle such as, but not limited to, linear and angular velocities, heading, and motion prediction are provided to a Bayesian estimation algorithm, and a final pose is computed.