RLWRLD picked for NVIDIA Korea AI investment, unveils DexBench

RLWRLD's selection for NVIDIA's investment plan follows a June 2026 collaboration between the two companies on common evaluation standards and data standards for dexterous manipulation. That earlier work now has its first major output in DexBench, which reproduces 15 tasks across retail, logistics, assembly, and hotel food service on NVIDIA Isaac Lab-Arena. The benchmark is built on NVIDIA's open-source robot evaluation framework and the open-source physics engine Newton, and it assesses not only task success rates but also generalization to unfamiliar objects and physical properties. The investment is part of NVIDIA's broader plan for Korean AI startups, announced at a forum co-hosted with Korea's Ministry of Science and ICT. RLWRLD intends to combine NVIDIA's simulation technology with its own RLDX-1 robotics foundation model and to expand sim-to-real validation, robot platforms, and task types. Industrial partners such as Genesis AI and Mimic Robotics are already testing DexBench through early access.