Lidar ICP Motion Estimation for Stable HD Map Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques for aligning point clouds from lidar sensors on moving vehicles are unstable and prone to error due to incorrect motion estimation, leading to inefficiencies and inaccuracies in high-definition map generation for autonomous navigation.
Innovation Solution
The implementation of a kinematic iterative closest point (ICP) technique with integrated motion estimation, which adjusts velocity parameters to optimize the alignment of point clouds by determining correspondences and transforming points based on their capture times, using non-linear optimization and physical constraints to prevent divergence.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional ICP techniques are used to align point clouds, then the alignment process can be performed, but the process becomes unstable and prone to error due to incorrect motion estimation
Solution Approach 1:
The patent applies preliminary action by performing motion estimation using odometry data before the ICP alignment process. This pre-estimation of vehicle motion (position, orientation, velocity) allows the system to pre-compensate for vehicle movement effects on the point cloud data, providing a more accurate initial state for the ICP algorithm and preventing divergence during iterations.
Solution Approach 2:
The patent introduces an intermediary component - the odometry-based motion estimation system - that mediates between the raw sensor data and the ICP alignment process. This intermediary provides corrected motion parameters that account for vehicle dynamics, enabling more accurate transformation of point cloud coordinates and improving overall alignment reliability.
2Measurement precision
If conventional ICP techniques iterate multiple times to improve alignment, then alignment accuracy may improve, but the computational process becomes inefficient and may diverge resulting in larger errors
Solution Approach 1:
By performing motion estimation before ICP alignment, the patent reduces the number of iterations needed for convergence. The pre-compensated point cloud data starts closer to the final aligned state, allowing the ICP algorithm to converge faster with fewer iterations, thus improving computational efficiency while maintaining or enhancing alignment accuracy.
Solution Approach 2:
The patent implements feedback by using the motion estimation results to continuously refine the alignment process. The odometry data provides ongoing corrections to the point cloud transformations, enabling the system to maintain accurate alignment with fewer iterations and preventing the divergence that would otherwise require additional corrective iterations.
3Ease of manufacture
If conventional ICP techniques treat each point cloud as having a single center, then the processing is simplified, but the measured point cloud from lidar is not consistent with the real world
Solution Approach 1:
The patent applies segmentation by dividing the point cloud processing into multiple segments - each point is associated with a specific timestamp and vehicle pose. Instead of treating the entire point cloud as a single entity with one center, the system processes points in temporal segments, applying appropriate motion corrections to each segment based on the vehicle's state at the time of capture, thereby maintaining both processing manageability and real-world consistency.
Solution Approach 2:
The patent introduces dynamics by making the point cloud center and transformation parameters time-dependent. Each point cloud is associated with a sequence of vehicle poses over time, and the transformation applied to each point depends on its capture timestamp. This dynamic approach allows the system to account for continuous vehicle motion while maintaining computational feasibility through efficient temporal processing.
Data Source
AI summary
A system align point clouds obtained by sensors of a vehicle using kinematic iterative closest point with integrated motions estimates. The system receives lidar scans from a lidar mounted on the vehicle. The system derives point clouds from the lidar scan data. The system iteratively determines velocity parameters that minimize an aggregate measure of distance between corresponding points of the plurality of pairs of points. The system iteratively improves the velocity parameters. The system uses the velocity parameters for various purposes including for building high definition maps used for navigating the vehicle.