Physical artificial intelligence (AI) corporations RLWRLD will work with Korea's simulation and Robotics startups to advance a humanoid fine dexterity evaluation benchmark.
On the 7th, RLWRLD said it signed separate memorandums of understanding (MOUs) with PhysicsSimLab, SpaceAI, and NdotLight. Through this collaboration, RLWRLD plans to advance "DexBench," a humanoid fine dexterity evaluation benchmark under development with Nvidia.
DexBench is a benchmark that evaluates whether robots in real industrial settings can use hands like humans to perform complex tasks. Based on 18 tasks and 80 cases, it quantitatively verifies robots' precision manipulation capabilities in both simulation and physical environments. It is being developed as an open-source project integrated with the Nvidia Isaac Lab-Arena environment.
Lotte, SK Telecom, CJ Logistics, Hyosung, HL Mando, Fuji, All Nippon Airways (ANA), and Mitsui Chemicals are also participating as core partners of DexBench. Because building sophisticated three-dimensional (3D) assets, physics-based data, and a simulation validation framework is necessary, RLWRLD plans to form a "K-physical AI coalition" with Korea-based startups that have technology in these fields to advance the related technologies.
RLWRLD previously released as open source its self-developed Robotics foundation model (RFM) "RLDX-1." The company said the model achieved state-of-the-art performance across eight simulation benchmarks.
Ryu Jung-hee, CEO of RLWRLD, said, "For humanoids to be useful in real industrial settings such as manufacturing and logistics, replicating human-level fine dexterity is key," and added, "This collaboration is a meaningful starting point where a coalition of Korean startups, each with expertise in their fields, put their heads together to solve high-difficulty challenges needed for global industry standards."