Autonomous Robot Position Correction Using Style-Transferred Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous robots face performance degradation due to measurement and control errors when transitioning from a simulated environment to a real environment, leading to inaccurate position estimation and map generation.
Innovation Solution
A position correction method for autonomous robots using a style-transfer model to minimize noise by generating and correcting robot viewpoint and global maps, employing a camera, traveling distance sensor, and a control unit to estimate and correct positions based on style-transferred maps.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If the autonomous robot is learned on a simulator using 3D scanner data, then the robot can be trained efficiently in a controlled environment, but position estimation accuracy deteriorates when applied to real environments due to measurement errors and control errors
Solution Approach 1:
The patent introduces a correction unit that acts as an intermediary between the simulator-based robot and the real environment. This correction unit processes the robot viewpoint map and global map to calculate position correction values, thereby mediating the transition from simulated to real-world operation and improving position estimation accuracy without sacrificing training efficiency
Solution Approach 2:
The patent replaces direct mechanical measurement systems with a map-based correction system. Instead of relying solely on physical sensors and mechanical odometry, the system uses image processing and map matching algorithms to substitute and correct positional information, thereby improving accuracy while maintaining the simplicity of the original system
2Reliability
If noise modeling is applied to minimize the difference between simulator and real environment, then performance degradation is reduced, but the complexity of understanding and modeling the deployment environment increases
Solution Approach 1:
The patent implements a self-service correction mechanism where the robot autonomously generates its own correction data during operation. The correction unit automatically processes the robot viewpoint map and global map to calculate position corrections without requiring external intervention or complex pre-modeling of the deployment environment, thereby maintaining reliability while reducing complexity
3Measurement precision
If actual measurement data is collected during driving to correct position estimation, then position accuracy improves, but the system requires additional sensors and measurement infrastructure
Solution Approach 1:
The patent creates a virtual copy of the environment through map generation from camera images, and uses this copied representation for position correction. Instead of requiring additional physical sensors, the system generates and processes virtual maps (robot viewpoint map and global map) that serve as substitutes for direct measurement data, thereby improving accuracy without increasing hardware complexity
Data Source
AI summary
An autonomous driving robot includes a driving unit that moves the autonomous robot; a camera; a traveling distance measurement sensor; and a control unit that estimates a location of the autonomous robot using a captured image and traveling distance information. In this case, the operation control program generates a robot viewpoint map based on the image captured by the camera, estimates a location of the autonomous robot based on the robot viewpoint map and the measured traveling distance information, and generates a global map based on the robot viewpoint map and position estimation information, and the operation control program inputs the generated robot viewpoint map and global map into a style-transfer model, and inputs a style-transferred robot viewpoint map and a style-transferred global map output by the style-transfer model into the operation agent to correct the estimated position.