Floor Obstacle Detection Method Using Laser Distance Sensors

Floor Obstacle Detection Technology Using Laser Range Sensors via One-Class Classification Learning

This technology detects small floor obstacles by training a one-class classification model on data from a normal driving surface, obtained via a tilted 2D laser range sensor, and statistically determining whether real-time sensing data deviates from the normal range.

Existing grid map methods struggle to detect small obstacles due to the resolution limitations of 2D laser sensors, while learning-based techniques require large-scale training datasets that include obstacle data and often fail to identify minor protrusions due to sensor bias errors.

This technology proposes a method that converts data collected from normal driving surfaces into Mahalanobis distance-based feature data, registers it to a one-class classification model, and corrects sensor bias errors using a Kalman filter before applying real-time data to the model. It can be applied to cleaning robots and indoor delivery robots to accurately detect small dropped objects and thresholds without the need for obstacle data collection.

‍

‍

Key Features:

• Generating multiple normal floor characteristic data points through driving on a normal surface
• Registering the normal floor characteristic data into a pre-registered one-class classification model
• Acquiring sensing values from a laser range sensor as the mobile robot moves
• Detecting obstacles by generating floor characteristic data from sensing values and determining whether they are normal

‍

This invention was developed with support from the Ministry of Science, ICT and Future Planning for the development of commercial-grade autonomous driving controllers for unmanned transport robots in diverse environments; the Small and Medium Business Administration for integrated driving control systems for multiple intelligent autonomous transport robots; and the Ministry of Science, ICT and Future Planning for intelligent growth-type autonomous driving systems for unmanned vehicles operating safely in congested residential road environments.