South Korean startup RLWRLD has partnered with a luxury hotel and a logistics firm to capture real workplace expertise.

The staff wear body cameras while performing various tasks to help generate data for future robot training.

Building on this data, RLWRLD recently unveiled RLDX-1, a dexterity-focused foundation model designed for complex real-world industrial robotics tasks.

The system integrates the full robotics pipeline, including scalable data collection, advanced architecture design, training methods, and optimized deployment strategies.

The company claims RLDX-1 delivers state-of-the-art performance across both simulated environments and physical industrial settings, marking a step toward more capable robotic manipulation systems.

Dexterity-first robotics leap

The Korean firm’s RLDX-1 is a dexterity-first foundation model designed for high-precision robotic manipulation using high-degree-of-freedom (DoF) hands. The model targets complex real-world industrial tasks that require fine motor control, contact awareness, and long-horizon decision-making.

RLWRLD says current foundation models struggle with key capabilities such as context memory and force sensing. RLDX-1 addresses this through a full-stack robotics pipeline combining scalable data collection, a multi-stream transformer architecture, and integrated training and deployment strategies. The system is built around the Multi-Stream Action Transformer (MSAT), which processes vision, motion, memory, and torque signals in separate streams before fusing them for action generation, reports The Robot Report (TRR).

According to the firm, the model also incorporates a robotics-specialized vision-language model, motion and physics modules, and a cognition interface that compresses perception into memory tokens for long-term task tracking. A synthetic data engine and human hand motion capture pipeline expand training coverage for dexterous manipulation.

RLWRLD reports that RLDX-1 achieves state-of-the-art performance across simulated and real-world benchmarks, outperforming leading vision-language-action models in spatial, temporal, and contact-rich tasks. The system is designed to generalize across embodiments, including single-arm, dual-arm, and humanoid robots, enabling more adaptive and capable robotic manipulation in industrial environments, according to TRR.

Robot skills database

RLWRLD is also building a large-scale database of human workplace skills to train robots to perform complex real-world tasks. Workers in industries such as hospitality, logistics, and retail are being recorded by body-mounted cameras as they carry out routine yet highly skilled tasks, including food service preparation, warehouse operations, and retail organization.

At Lotte Hotel Seoul, staff movements during tasks like folding banquet napkins and table preparation are being captured in detail and converted into training data. Similar motion-tracking systems are being deployed at logistics facilities operated by CJ Logistics and retail environments run by Lawson. The aim is to capture fine-grained human dexterity, including grip, force, and sequencing of actions, reports The Asahi Shimbun (TAS).

RLWRLD is developing an AI software layer intended to power a new generation of robots that can operate across factories, warehouses, and eventually household environments. These systems, often referred to as physical AI, combine perception, decision-making, and physical interaction in real-world settings.

The company’s approach reflects a broader industry shift toward humanoid robotics, where replicating human hand dexterity is considered essential for practical deployment. South Korea is positioning itself as a competitor in this field, leveraging its strong manufacturing base and skilled workforce to generate high-quality training data for robotics, reports TAS.

National initiatives are supporting the push, including government programs to digitize expert skills for AI manufacturing systems. The government also recently launched a $33 million project to capture the know-how of master technicians in a database, aiming to improve productivity and address an aging, shrinking workforce through robotics.