Indoor Semantic Mapping Using Wall Corners for Stable Robot Navigation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional map construction methods for robot platforms, such as topological or grid maps using laser SLAM, lack semantic information, and visual SLAM is prone to errors due to light conditions and texture scarcity, making it difficult to create accurate indoor two-dimensional semantic maps, especially with dynamic environments.
Innovation Solution
A method combining laser SLAM with a deep learning model and visual sensors to extract wall corner semantics, using a lidar sensor and depth camera to construct a grid map, perform target identification, and fuse information using Bayesian estimation for incremental mapping, resulting in a richer semantic map for navigation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If visual SLAM is used to obtain environmental semantics, then rich texture information can be acquired, but the system becomes highly susceptible to light conditions and accumulates errors in map construction
Solution Approach 1:
The patent combines laser SLAM with visual semantic segmentation to merge the advantages of both methods. The laser SLAM provides accurate geometric mapping and pose estimation, while the visual semantic segmentation adds environmental semantics. This hybrid approach resolves the contradiction by using the reliable laser-based positioning to compensate for visual SLAM's error accumulation while retaining the semantic information extraction capability of visual methods.
Solution Approach 2:
The patent introduces an intermediary deep learning model (semantic segmentation model) that processes visual data to extract semantic information without relying on visual SLAM's pose estimation. This intermediary component allows the system to acquire environmental semantics through image processing while using laser-based methods for accurate positioning and map construction, thereby avoiding the reliability issues of visual SLAM.
2Measurement precision
If traditional laser SLAM is used to construct grid map, then topological and geometric information can be obtained, but semantic information of the environment cannot be extracted
Solution Approach 1:
The patent merges traditional laser SLAM with visual semantic segmentation. The laser SLAM component maintains its strength in providing precise topological and geometric information, while the added visual semantic segmentation component extracts semantic information from images. This combination resolves the contradiction by preserving the accurate geometric mapping capability of laser SLAM while supplementing it with semantic content from visual data.
Solution Approach 2:
The patent segments the map construction process into two independent parts: geometric mapping (handled by laser SLAM) and semantic extraction (handled by visual semantic segmentation). This segmentation allows each component to specialize in its strength - laser SLAM for accurate geometric information and deep learning models for semantic information - thereby resolving the contradiction without compromising either aspect.
3Loss of information
If visual SLAM is used in dark environments with fewer textures, then semantic information can be obtained, but the boundary becomes unclear and operation load increases
Solution Approach 1:
The patent introduces an intermediary semantic segmentation model that processes visual data independently of the SLAM system. This intermediary component extracts semantic information from images using deep learning, which is more robust to lighting conditions and texture scarcity than traditional visual SLAM. The semantic segmentation model operates as a separate module, managing its own computational load and providing semantic information without increasing the overall system complexity.
4Adaptability or versatility
If objects in the environment move, then the environment changes dynamically, but the constructed semantic map experiences large changes and becomes unstable
Solution Approach 1:
The patent implements a dynamic semantic map construction approach where the semantic map is continuously updated as the robot explores the environment. The system uses incremental mapping techniques that allow the semantic map to adapt to dynamic changes while maintaining stability through consistent update rules. This dynamic approach resolves the contradiction by enabling the map to evolve with the environment while preserving structural integrity through systematic update procedures.
Data Source
AI summary
A method of constructing an indoor wall corner two-dimensional semantic map based on a robot platform includes controlling, by the master control computer, the robot platform to move indoors, collecting, by the lidar sensor, a distance between an indoor object and the robot platform and a direction angle between the indoor object and the robot platform in real time and transmitting the distance and the direction angle to the master control computer, and obtaining, by the master control computer, an environment grid map and a real-time pose of the robot platform by processing the distance between the indoor object and the robot platform and the direction angle between the indoor object and the robot platform based on Gmapping algorithm; constructing a semantic segmentation data set as a training set on the master control computer and obtaining a predicted non-wall corner semantic label.