The Importance of Robotic Testbeds for Real-World Data
2025.11.28 NAVER LABS
In the AI race, the focus of competition is shifting from models to data, and this is also true of robotics research: beyond simulations based on synthetic data, true technological competitiveness now depends on how precisely a system can understand and reflect real environments. The essential foundation for this is the robot testbed—a real-world environment where robots can be tested and operated to accumulate data. Testbeds allow the continuous advancement of technology by collecting high-quality data in the physical world, training models with it, and then validating them again in real spaces.
Digital twins that precisely replicate physical spaces are paired with high-quality spatial data to form the starting point for robots to perceive their surroundings and build simulations. When combined with the driving and visual data gathered as robots move through and experience real environments, the robots’ navigation becomes more efficient, and their ability to respond to surrounding objects, people, and situations becomes even stronger. Furthermore, user and operational data accumulated through the real deployment of robot services become critical assets for refining service quality, improving efficiency, and enhancing user satisfaction.
NAVER LABS recognized early on the importance of real-world data and has directly designed and operated every step, from developing its own mapping devices to building robots and conducting real-environment testing. This approach has enabled NAVER LABS to establish a virtuous cycle of “data collection → learning and application → real-world validation → iterative improvement.”
Virtuous Cycle of Real Data through Testbeds
Precision Mapping Technology that Replicates the Real World
NAVER LABS has developed its own series of mapping devices to accurately digitize physical spaces. Designed to collect spatial data effectively across diverse environments and conditions, the lineup consists of four specialized series.
These solutions digitize real-world environments with high precision, enabling robots to perceive their surroundings accurately. The collected data is integrated into NAVER’s Digital Twin platform and the ARC Eye cloud positioning system, where it is expanded into various robot services and further used as training data for robotics models.
M Series: High-precision mapping robots that generate 3D digital twins of indoor spaces
T Series: Wearable mapping devices that collect data while navigating complex indoor and outdoor terrains, such as stairways
R Series: Mobile mapping systems that capture road environments for HD map creation
P Series: Panoramic mapping systems that collect nationwide street-level data
Mapping Robots and Devices for Diverse Environments
Real-World Robotics Validation Data: From Hardware to Services
Since 2017, NAVER has continuously advanced its technology by operating robots directly in real-world environments. The core principle is to validate and refine robots and services through continuous deployment in actual settings. Representative examples include the AROUND series used in cafés, libraries, and logistics sites; the Mini-Cheetah, a quadruped robot co-developed with MIT; AMBIDEX, a dual-arm collaborative robot; and Rookie, the service robot currently operating inside 1784, NAVER’s second headquarters. In addition, autonomous delivery robots, warehouse automation robots, and autonomous shuttles have all accumulated real-environment data to advance their performance.
Everything a robot sees, experiences, and encounters while moving becomes data that enhances driving efficiency and safety. Real-world information, such as human movement patterns, types of obstacles, and spatial patterns, helps robots become increasingly smarter and more adaptable.
NAVER also designs the entire service operation surrounding its robots as user behavior data, satisfaction levels, and service efficiency metrics gathered through real operation serve as a foundation for refining service quality. NAVER directly oversees planning, operation, and maintenance for its robot delivery services, which are integrated with NAVER Order, NAVER Pay, and POS systems to ensure greater reliability and a more fine-tuned result.
NAVER’s Large-Scale Robot Testbeds
Regarded as a model for next-generation smart buildings, NAVER’s second headquarters, 1784, is a “tech convergence building” that integrates robotics, autonomous driving, AI, 5G, and cloud technologies. At 1784, the service robot Rookie travels alongside people using standard elevators and moves vertically through Roboport, an elevator dedicated to robots. Through this setup, NAVER demonstrates the efficiency required to operate large-scale robot services reliably in high-rise environments.
Vertical mobility testing of robots inside NAVER 1784
Meanwhile, NAVER’s hyperscale data center, Gak Sejong, serves as a testbed where cloud, robotics, and AI technologies are integrated. Through robotic automation deployed in real operational settings, the center maximizes efficiency from the intake of IT assets to ongoing management. Across a vast campus equivalent to 41 soccer fields in length, multiple robots travel between buildings, transporting materials and supporting human workflows. This system functions almost like a horizontally deployed elevator network.
While 1784 functions as a testbed for smart buildings, Gak Sejong serves as a testbed for smart campuses. These two infrastructures together form the foundation for advancing NAVER’s robot AI technologies.
Robots operating across diverse indoor and outdoor environments at Gak Sejong
Synthetic data and simulators are powerful tools for advancing robot AI. However, they cannot fully reproduce the complexity of real-world environments, nor can they replace the authenticity revealed through service data. NAVER continues to experiment with and refine its robotics technologies in actual physical spaces and through real services, using data gathered directly from where robots and users interact.