Grid-Based Mobile Body Localization With Linear Complexity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Grid-based Global Localization methods face exponential computational complexity, limiting their use to small grids due to the need for extensive calculations in updating pose probabilities, especially when dealing with large environments and noisy observations.
Innovation Solution
Transforming the Global Localization problem into a mapping problem by identifying border cells in the occupancy grid as 'mapping sources' that use sensor observations to update pose probabilities in position grids, reducing complexity through Bayesian fusion and integer arithmetic for efficient computation on low-power devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If grid-based Global Localization methods are used to determine robot pose in a known map, then robustness and ability to deal with sensor uncertainty are improved, but computational complexity increases exponentially with grid size
Solution Approach 1:
The patent segments the global localization problem into multiple local localization problems by dividing the pose space into smaller grids. Each local grid maintains pose probability distributions independently, avoiding the exponential complexity of computing over the entire large grid while preserving robustness through multiple localized estimations.
Solution Approach 2:
The patent transforms the computational problem by changing the dimension of operation from the entire pose space to smaller subspaces. By organizing computation in a hierarchical structure with multiple levels of grids, the method reduces the effective dimensionality that needs to be processed at each computational step, thereby reducing exponential complexity.
2Area of stationary object
If large grids are used to cover large environments, then localization coverage is improved, but computational cost becomes intractable
Solution Approach 1:
The patent divides the large environment into multiple smaller grids that can be processed independently and in parallel. This segmentation allows the system to cover large areas by combining results from multiple small grids, achieving large-area localization without the intractable computational cost of a single large grid.
Solution Approach 2:
The patent computes pose probabilities only for relevant regions within each local grid rather than exhaustively processing the entire pose space. By focusing computational resources on partial regions that are most likely to contain the robot, the method achieves efficient localization across large environments without intractable computational requirements.
Data Source
Figure 1~3
Figure 4~6B
Figure 6C~7
AI summary
A method of localizing a mobile body (MB) in a known environment, comprising the following steps: a) defining an occupancy grid (G) modeling the environment; b) defining a set of position grids (II) each position grid being associated to a heading of the mobile body; c) receiving a time series of measurements (Z1, Z2,...) from a distance sensor carried by the mobile body; and d) upon receiving a measurement of the time series, updating the pose probabilities of the position grids as a function of present values of said occupancy probabilities and of the received measurement; wherein step d) is carried out by applying an inverse sensor model to the received measurement, while considering the distance sensor co-located with a detected obstacle and by applying Bayesian fusion to update the pose probabilities of the position grids.