Physical AI company RLWRLD has unveiled its self-developed robotics foundation model 'RLDX-1', designed to implement human-level precision in robotic hands. At the 'Dexterity Night in Seoul' event held on June 10, a humanoid robot equipped with RLDX-1 successfully picked up a wireless mouse with five fingers and placed it into a box.
RLWRLD CEO Ryu Joong-hee stated, "Industrial sites still require human hands for certain tasks," emphasizing the goal of automating precise processes that demand human dexterity. "Even the most automated factories in Korea only reach 75% automation. The remaining 25% is still handled directly by people."
What sets RLDX-1 apart from existing vision-language-centric VLA models is its integration of vision, language, torque, tactile feedback, and task memory within a single model. The company has formed partnerships with major firms including SK Telecom, LG Electronics, CJ Logistics, Lotte, KDDI, and ANA, and is collaborating with Nvidia on DexBench, a universal benchmark for evaluating humanoid robot hand manipulation.
In both simulation and real-robot tests, RLDX-1 outperformed all competing models, including Nvidia GR00T. It was the first robotics foundation model to exceed 78% on a simulation benchmark and recorded more than 10 times GR00T's success rate on precise manipulation tasks requiring human dexterity.
