Radar Pose Estimation Using Hyper-Local Submaps and Multi-View ICP
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current radar pose estimation methods for autonomous vehicles, particularly using low-cost signal system on chip (SoC) based millimeter wave radars, face challenges with noisy and sparse radar point clouds, which affect the accuracy and robustness of dynamic calibration.
Innovation Solution
The system employs an automated driving controller that determines a hyper-local submap based on consecutive aggregated filtered data point cloud scans and associated pose estimates, using iterative closest point (ICP) alignment and multi-view non-linear ICP algorithms to adjust poses, and executes loop detection and non-linear optimization routines to refine the radar pose estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If low-cost signal system on chip (SoC) based millimeter wave radar is used for autonomous vehicles, then cost is reduced, but measurement precision deteriorates due to noisy and sparse radar point clouds
Solution Approach 1:
The patent combines multiple radar point cloud scans into a unified representation by identifying corresponding points across scans and merging their coordinates. This aggregation process consolidates noisy individual scans into a more reliable composite point cloud, improving measurement precision while maintaining the use of low-cost radar sensors.
Solution Approach 2:
The patent introduces an automated driving controller as an intermediary that processes raw radar point cloud data through multiple stages: initial pose estimation, loop detection, and pose graph optimization. This intermediary system transforms noisy radar measurements into accurate pose estimates, bridging the gap between low-cost sensors and high-precision requirements.
2Measurement precision
If manual recalibration is performed when sensors are moved or vehicle undergoes wheel alignment, then measurement precision is restored, but loss of time increases due to cumbersome manual process
Solution Approach 1:
The patent implements self-service calibration through automated loop detection and pose graph optimization. The system automatically detects when the vehicle returns to a previously visited location, identifies the loop closure, and adjusts the pose graph accordingly. This eliminates the need for manual recalibration operations while maintaining measurement precision.
Solution Approach 2:
The patent employs feedback mechanisms where the system continuously monitors radar point cloud data, detects loops, and adjusts pose estimates in real-time. This closed-loop feedback system automatically corrects calibration drift caused by sensor movement or wheel alignment, restoring measurement precision without manual intervention and reducing recalibration time to near zero.
Data Source
AI summary
A scan matching and radar pose estimator for determining a final radar pose for an autonomous vehicle includes an automated driving controller that is instructed to determine a hyper-local submap based on a predefined number of consecutive aggregated filtered data point cloud scans and associated pose estimates. The automated driving controller determines an initial estimated pose by aligning a latest aggregated filtered data point cloud scan with the most recent hyper-local submap based on an iterative closest point (ICP) alignment algorithm. The automated driving controller determines a pose graph based on the most recent hyper-local submap and neighboring radar point cloud scans, and executes a multi-view non-linear ICP algorithm to adjust initial estimated poses corresponding to the neighboring radar point cloud scans in a moving window fashion to determine a locally adjusted pose.