Mobile Robot LiDAR Localization for Partial-View Target Objects
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for localizing mobile robots with respect to large objects, such as aircraft, face challenges in accuracy due to reliance on feature extraction and machine-learning techniques, which are computationally expensive and limited by the need for whole-object visibility, restricting rapid deployment and varying object shapes/sizes.
Innovation Solution
A method using LiDAR data to create an initial map of a reference object, iteratively updating pose data, extracting relevant LiDAR points, and determining localization pose based on confidence values, while filtering out non-target object data points to enhance accuracy and adaptability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If feature extraction is used to localize the robot with respect to the target object, then localization accuracy may be improved, but computational cost increases significantly
Solution Approach 1:
The patent extracts only the relevant subset of LiDAR data points that correspond to features of the target object from the complete point cloud representation, rather than processing all data points. This extraction is performed based on the estimated pose of the mobile robot, allowing the system to focus computational resources on the most informative data points for localization, thereby reducing overall computational cost while maintaining localization accuracy.
2Productivity
If machine-learning techniques are used to directly detect the object in sensor data, then localization speed may be improved, but the requirement for whole-object visibility restricts applicability to varied object shapes and sizes
Solution Approach 1:
The patent segments the target object into multiple parts by processing the point cloud data in sequential sampling instances as the robot traverses the environment. Instead of requiring the entire object to be visible at once, the system builds up a complete localization solution by integrating information from multiple partial views collected over time, enabling rapid deployment across objects of varied shapes and sizes.
Solution Approach 2:
The patent employs dynamic pose estimation by iteratively updating the robot's pose based on new LiDAR data in each sampling instance. The estimated pose from the previous instance is used to extract relevant features, which then inform the pose update in the current instance. This dynamic approach allows the system to adapt to objects of varying shapes and sizes without requiring pre-learning, maintaining both speed and versatility.
3Use of energy by moving object
If the robot computes pose once and extrapolates future poses without sensor feedback, then computational cost is reduced, but localization accuracy deteriorates over time
Solution Approach 1:
The patent implements continuous feedback-based pose correction by comparing the extracted LiDAR data points from the environment with the expected points from the initial map of the target object. The discrepancies between observed and expected features are used to correct the estimated pose in each sampling instance, preventing drift accumulation and maintaining high localization accuracy over time while using computational resources efficiently through selective feature processing.
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 improves localization accuracy and adaptability by focusing on relevant LiDAR data points, maintaining high localization precision even with limited data points and varying object shapes, enabling robust and efficient object-based localization.
Implementation Method 1
obtaining LiDAR data of the target object captured by a LiDAR device of the mobile robot
Data Source
AI summary
A method of localizing a mobile robot 100 with respect to a target object 601 using an initial map of a reference object which is a representation of the target object is disclosed herein. The method comprises obtaining LiDAR data 503 of the target object 601 captured by a LiDAR device 203 of the mobile robot 100 as the mobile robot 100 traverses in an environment associated with the target object 601 the LiDAR data including respective point cloud representations of the target object 501 and the environment in various sampling instances; iteratively updating pose data representing estimated pose of the mobile robot 100 with respect to a known reference location in corresponding sampling instances as the mobile robot traverses the environment; extracting a subset of LiDAR data points in the point cloud representation of a previous sampling instance, the subset of LiDAR data points corresponding to features of the target object 601 in the initial map based on the estimated pose of the mobile robot 100 associated with the previous sampling instance; obtaining desired LiDAR data points in the point cloud representation in a current sampling instance, the desired LiDAR data points including current LiDAR data points which correspond to the extracted subset of LiDAR data points in the previous sampling instance; and determining a localization pose of the mobile robot 100 with respect to the target object 601 in the current sampling instance based on the desired LiDAR data points. A system for localizing a mobile robot is also disclosed herein.