Vehicle Pose Estimation Using LiDAR Point Clouds and UKF
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing navigation systems for autonomous vehicles face accuracy degradation in complex urban environments due to poor GPS satellite signals, leading to positioning errors and global inconsistency in high-resolution maps.
Innovation Solution
The implementation of a navigation system that combines LiDAR, GPS, IMU sensors, and camera data with Unscented Kalman Filter (UKF) and Normal Distributions Transform (NDT) for improved vehicle pose estimation, using trajectory interpolation and pose graph optimization to reduce error accumulation and enhance positioning accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If GPS, IMU, and wireless base stations are used for vehicle pose estimation, then positioning accuracy of about 10 cm is achieved in normal conditions, but positioning accuracy is greatly degraded in complex urban environments with poor GPS satellite signals
Solution Approach 1:
The patent combines multiple sensors (LiDAR, GPS, IMU, cameras) to capture features and aggregate point cloud frames. By merging data from these diverse sensors, the system achieves robust positioning that maintains accuracy in complex urban environments where GPS alone fails, directly resolving the contradiction between normal-condition accuracy and urban-environment reliability
Solution Approach 2:
The patent introduces point cloud frames and pose graph optimization as intermediary elements between the sensors and the final positioning result. These intermediaries process and integrate sensor data to produce accurate pose estimates even when GPS signals are poor, thereby maintaining reliability in complex environments
2Measurement precision
If real-time point clouds are introduced to improve positioning accuracy, then positioning accuracy is enhanced, but error accumulation still occurs due to poor GPS satellite signals
Solution Approach 1:
The patent implements pose graph optimization that uses feedback from multiple point cloud frames to correct positioning errors. The system continuously compares estimated poses with observed point cloud features and adjusts the pose estimates to minimize errors, preventing error accumulation even when GPS signals are poor
Solution Approach 2:
The patent performs preliminary pose estimation using point cloud frames before final map aggregation. By pre-processing and optimizing poses using point cloud data and pose graph methods, the system eliminates positioning errors in advance, preventing their accumulation in the final map product
3Manufacturing precision
If high-resolution maps are generated by aggregating multiple point cloud frames based on 3-D pose information, then map resolution is improved, but global inconsistency occurs due to positioning errors
Solution Approach 1:
The patent uses pose graph optimization with feedback loops to ensure global consistency. The system continuously refines pose estimates by comparing point cloud features across multiple frames and adjusting poses to maintain consistency, allowing high-resolution map aggregation without global inconsistency
Solution Approach 2:
The patent performs preliminary pose optimization and error elimination before aggregating point cloud frames for map generation. By pre-processing the pose data to eliminate errors and ensure consistency, the system enables high-resolution map aggregation while maintaining global consistency throughout the final map product
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach provides high-accuracy vehicle positioning and point cloud data aggregation, reducing error accumulation and ensuring consistent map generation even in challenging environments.
Implementation Method 1
a LiDAR radar, to capture features of the road on which the vehicle is driving or the surrounding objects
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Embodiments of the disclosure provide systems and methods for positioning a vehicle. The system includes a communication interface (202) configured to receive point cloud frames with respect to a scene and initial pose data of a vehicle captured by sensors (150) equipped on the vehicle as the vehicle moves along a trajectory. The system also includes a storage (208) configured to store the point cloud frames and the initial pose data. The system further includes a processor (204) configured to estimate pose information of the vehicle associated with each of the point cloud frames based on the initial pose data and the point cloud frames. The processor (204) is also configured to adjust the estimated pose information of the vehicle based on a model. The model includes a spatial relationship and a temporal relationship among the plurality of point cloud frames. The processor (204) is further configured to position the vehicle based on the adjusted pose information.