Keimyung University Develops Core SLAM Technology for Quadruped Robots Using Limited Field-of-View LiDAR

Significant Improvement in Autonomous Navigation Accuracy Through Terrain Recognition Based on Walking Data
Published in a Prestigious Q1-Level SCIE Journal
Achievement of Industry-Academia Collaboration with Hanwha Systems

The research team led by Professor Seonghoon Hong from the Department of Robotics Engineering at Keimyung University (President Shin Ilhee) has developed a core technology for simultaneous localization and mapping (SLAM) in quadruped robots using LiDAR sensors with a limited field of view, and published their findings in a prestigious international journal.

This research was published in "Robotics and Autonomous Systems," a Q1-level SCIE journal in the fields of automation and control systems, computer science, and artificial intelligence.

(From left) Seonghoon Hong, Professor at Keimyung University; Seokmin Yoon, Senior Researcher at Hanwha Systems; Rowoon Lee, Master's graduate from Keimyung University (currently working at Riibotics).

The title of the paper is "Terrain-Based Place Recognition for LiDAR SLAM of Quadruped Robots with Limited Field-of-view Measurements."

LiDAR (Light Detection And Ranging) is a sensor technology that measures distance and position by emitting laser light and calculating the time it takes for the reflected light to return from objects.

It enables the recognition of the surrounding environment as three-dimensional point cloud data, making it a core technology for robot localization and mapping (SLAM).

It is also less affected by changes in lighting or dark environments compared to other sensors.

Recently, solid-state LiDAR sensors have been gaining attention for their advantages of being lightweight and low-power compared to omnidirectional rotating types, making them suitable for mobile robots such as quadruped robots.

However, due to their structurally limited field of view, there have been ongoing issues with decreased SLAM performance during place recognition and loop-closure processes.

To overcome these limitations, the research team focused on the force/contact information acquired during the walking process of quadruped robots.

They proposed a novel approach that reconstructs terrain surfaces using walking patterns and foot position data observed as the robot traverses characteristic terrains such as slopes or stairs, and utilizes this information for place recognition.

This is considered an alternative that can ensure stable SLAM performance even in environments with limited sensor fields of view.

This research was conducted through industry-academia collaboration with Hanwha Systems, with Rowoon Lee, a master's graduate from the Department of Robotics System Engineering at Keimyung University (currently working at Riibotics), participating as the first author.

Professor Seonghoon Hong served as the corresponding author, and Seokmin Yoon, Senior Researcher at Hanwha Systems, participated as a co-author.

Professor Seonghoon Hong stated, "This research complements the limitations of conventional sensor-centric SLAM by utilizing the unique walking information of quadruped robots and their interaction with terrain. It is particularly meaningful that a student who completed both undergraduate and master's programs at our university has published research results as the first author in a prestigious international journal."

This achievement is expected to have significant applicability in related research fields such as quadruped robots, LiDAR SLAM, limited field-of-view sensors, place recognition, and robot autonomous navigation, and is anticipated to contribute to the advancement of autonomy and environmental perception technologies for mobile robots in the future.