Glass wall detection without the use of expensive sensors DGIST lowers autonomous driving cost by 90%

□ A research team led by Professor Kyungjoon Park at the Department of Electrical Engineering and Computer Science, Daegu Gyeongbuk Institute of Science & Technology (DGIST; President Kunwoo Lee) has developed autonomous driving software that allows inexpensive sensors to detect transparent obstacles such as glass walls, providing an alternative to high-performance sensors. This technology can be used in existing robots, negating the need for additional equipment while ensuring detection performance that is equal to that offered by expensive conventional equipment.

□ Autonomous driving robots typically use LiDAR sensors to detect their surroundings and navigate. Functioning as “laser eyes,” expensive LiDAR sensors determine distance and structure by projecting light and measuring reflection time. Inexpensive LiDAR sensors cannot detect transparent objects such as those made of glass; they may mistake them for empty space, potentially resulting in a collision. High-resolution ultrasonic LiDAR sensors or cameras do not have this limitation, but their use increases system complexity and raises costs by hundreds of thousands to millions of won.

□ To provide an alternative, a DGIST research team led by Professor Kyungjoon Park developed probabilistic incremental navigation-based mapping (PINMAP), an algorithm that approaches problem-solving via software, not hardware. PINMAP accumulates rare point data that inexpensive LiDAR sensors can detect only sporadically. Using these data, PINMAP probabilistically calculates the likelihood of the presence of glass walls over time.

□ The PINMAP algorithm is based on Cartographer (map charting) and Nav2 (navigation), which are open-source tools that are widely used in the ROS 2 ecosystem. PINMAP has the advantage of easy applicability while eliminating the need to change the existing system structure. Instead of upgrading the sensors at a high cost, the algorithm alters the way the existing sensors handle data; that is, it uses software to improve the detection performance of inexpensive LiDAR sensors.

□ In a real-world experiment conducted at DGIST, PINMAP detected glass walls with 96.77% accuracy, which is well above the nearly 0% detection rate of the traditional approach using the same inexpensive LiDAR sensors (Cartographer-SLAM). The software difference that PINMAP offers demonstrated a tremendous performance boost.

□ Professor Park said, “PINMAP flips the conventional wisdom that hardware performance equals system performance and proposes a new standard whereby software can improve sensor capabilities. This study shows that ensuring stable autonomous driving is possible without relying on high-performance equipment.”

□ The algorithm the research team developed offers a substantial economic advantage because it achieves detection performance comparable to that of expensive LiDAR sensors at less than one-tenth of the cost. This technology is expected to reduce collisions between autonomous driving robots and glass or transparent acrylic walls in indoor spaces such as hospitals, airports, shopping malls, and warehouses, thus contributing to the large-scale deployment of service robots.

□ The study was funded by the Mid-Career Researcher Support Program under the National Research Foundation of Korea. The first author was Jiyoung Chae, a graduate student enrolled in a combined Master’s and PhD program in the Department of Electrical Engineering and Computer Science at DGIST. The research was published online on May 7, 2025 in IEEE Transactions on Instrumentation and Measurement, one of the most recognizable international journals. Professor Park, the corresponding author listed on the article, is the CTO at S-Innovations, a robotics software startup. Professor Park works diligently to overcome cost barriers to the large-scale deployment of robots at industrial sites.