Autonomous mobile robot and method for correcting its position

Position Correction Technology for Autonomous Robots Using Style Transfer Learning Models

This technology corrects sensor and control errors that occur during autonomous robot localization and mapping. It generates a noise-minimized map by inputting real-time robot-view maps and global maps into a style transfer learning model, which is then used to calibrate the robot's position.

Existing autonomous robots often suffer from degraded localization performance in real-world operation due to discrepancies between simulated and actual environments, as well as errors in odometry sensors and motor control.

This technology utilizes an operation control program to generate robot-view and global maps. By applying a style transfer learning model between ground-truth image sets and real-world image sets, it produces transformed map data to calibrate the navigation agent's position estimates, enabling precise localization and mapping in real-world environments. Since it bridges the gap between simulation and reality using only a learning model—without the need for additional sensors—it significantly reduces development costs and trial-and-error during the commercialization of logistics and service robots.

‍

‍

Key Features:

• A drive unit, camera, and odometry sensor for moving the autonomous robot
• A control unit that estimates the autonomous robot's position using captured video and distance data
• An operation control program that generates robot-view and global maps based on video captured at each time step via a navigation agent
• A configuration that calibrates position estimates by inputting the generated robot-view and global maps into a style transfer learning model

‍

This invention was developed with support from the Ministry of Science and ICT for learning to establish mid-to-long-term task plans for service robots through hierarchical understanding of 3D information.