Point Cloud Registration with Anchor Poses and ADMM for HD Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current point cloud registration algorithms for autonomous driving vehicles are heavily dependent on GPS signals, which can be error-prone and unreliable in environments with tall buildings or dense forests, and are computationally complex, leading to inaccuracies and inefficiencies in constructing high-definition maps.
Innovation Solution
The method selects high-confidence anchor poses to reduce the number of decision variables in point cloud registration, uses parallel processing with the regional iterative closest point (ICP) algorithm, and adds a regularity term to the cost function to improve alignment and connection between sub-maps, leveraging the alternating direction method of multipliers (ADMM) for optimization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If GPS signals are used for point cloud registration, then the registration process can be performed, but the accuracy deteriorates due to signal errors and environmental degradation
Solution Approach 1:
The patent introduces an intermediary optimization model that acts as a mediator between GPS-based initial poses and final registered point clouds. This model uses pose graphs and optimization algorithms to refine positions, filtering out GPS errors and environmental degradation effects while maintaining the benefits of GPS-based initialization.
2Productivity
If the map area is divided into sub-maps for parallel processing, then the computational complexity is reduced, but the connection between sub-maps becomes weak leading to constant offsets
Solution Approach 1:
The patent merges multiple sub-maps into a unified pose graph structure where overlapping regions are integrated through shared poses. The optimization model combines results from all sub-maps, strengthening connections between them and eliminating constant offsets while preserving parallel processing benefits.
Solution Approach 2:
The patent creates a universal optimization framework that handles both individual sub-map registration and inter-sub-map coordination through a unified cost function. This multi-functional approach simultaneously optimizes local alignment and global consistency, ensuring map stability while maintaining processing efficiency.
3Area of stationary object
If the ICP algorithm is applied to large map areas, then comprehensive coverage is achieved, but the computational time becomes excessive
Solution Approach 1:
The patent segments the large map area into multiple sub-maps that can be processed in parallel. Each sub-map is registered independently using the ICP algorithm, significantly reducing computational time. The segmentation is then integrated through the pose graph optimization to achieve comprehensive coverage without excessive processing time.
4Device complexity
If the number of decision variables is reduced by selecting anchor poses, then the computational complexity is lowered, but the flexibility of the registration process is constrained
Solution Approach 1:
The patent implements a dynamic approach where the number and selection of anchor poses can be adjusted based on the specific registration scenario. The optimization model adaptively determines the appropriate level of constraint, maintaining flexibility while managing computational complexity. This dynamic balancing allows the system to adapt to different map sizes, densities, and accuracy requirements.
Data Source
AI summary
In one embodiment, a system and method for point cloud registration of LIDAR poses of an autonomous driving vehicle (ADV) is disclosed. The method selects poses of the point clouds that possess higher confidence level during the data capture phase as fixed anchor poses. The fixed anchor points are used to estimate and optimize the poses of non-anchor poses during point cloud registration. The method may partition the points clouds into blocks to perform the ICP algorithm for each block in parallel by minimizing the cost function of the bundle adjustment equation updated with a regularity term. The regularity term may measure the difference between current estimates of the poses and previous or the initial estimates. The method may also minimize the bundle adjustment equation updated with a regularity term when solving the pose graph problem to merge the optimized poses from the blocks to make connections between the blocks.