#United States #New York #NVIDIA #RLWRLD #DexBench

RLWRLD and NVIDIA Collaborate to Set New Industry Standards for Humanoid AI

On June 9, 2026, Real World Inc. (RLWRLD), headquartered in the United States and led by CEO Ryu Jun-Hye, announced an exciting partnership with NVIDIA aimed at creating the next generation of standards for humanoid robot AI. This initiative focuses on building an industry-wide infrastructure essential for advancing dexterous manipulation capabilities, a crucial area for the practical application and commercialization of humanoid AI.

Key Objectives of the Partnership

The collaborative effort will center around three primary goals:
1. Developing a shared benchmark named DexBench, which will evaluate dexterous manipulation skills.
2. Establishing standardized data standards for fine motor skill learning.
3. Strengthening collaboration with NVIDIA's robotics development platform, Isaac Lab, and the Isaac Lab-Arena.

These targets aim to lay a comprehensive foundation for humanoid robots to efficiently perform intricate tasks—such as assembly, sorting, and packaging—highlighting the urgent need for common frameworks in the industry.

The Importance of Dexterous Manipulation

Dexterous manipulation is critical for humanoid robots, enabling them to execute delicate functions that are pivotal in various industrial tasks. However, a key challenge remains: there is currently a lack of standardized benchmarks to objectively assess these capabilities across the industry. This deficiency hinders technological advancements and broader commercial adoption of humanoid AI.

Ryu Jun-Hye, CEO of RLWRLD, emphasized that without a common language to evaluate and replicate the precision movements of robotic hands, the commercial potential of humanoid AI is constrained. Through this collaboration with NVIDIA, RLWRLD aims to establish DexBench and data standards to foster the necessary infrastructure that can propel the entire industry forward, ultimately setting a new global benchmark for humanoid AI ecosystems.

The Dual Validation Framework of DexBench

DexBench will be integrated with NVIDIA’s Isaac Lab-Arena environment, facilitating the evaluation and validation of dexterous manipulation abilities within both simulated and real-world settings. This benchmarking system will focus on 18 essential tasks within industrial environments and evaluate performance based on five key metrics:

• - Diversity of grips
• - Spatial precision
• - Temporal accuracy
• - Contact precision
• - Situational awareness

By developing standardized evaluation criteria based on real industrial tasks, DexBench will create a common ground for assessing humanoid robot performance, thus providing manufacturers, research institutions, and companies with a unified evaluation standard. This will support the pathway from benchmark assessments to tangible commercial applications.

Establishing Humanoid Data Standards for Global Manufacturers

In collaboration with NVIDIA, RLWRLD is also focusing on standardizing data formats that can be utilized in training humanoid robots for dexterous operations, ensuring compatibility with the Isaac Lab pipeline. This framework is intended to serve as a universal data interface for robotics manufacturers and research institutions worldwide.

The RLDX-1 model, designed for humanoid dexterous manipulation, has shown outstanding performance across established simulation benchmarks, greatly outperforming leading models from NVIDIA. These outcomes not only validate RLWRLD's architectural approach but also present key challenges in establishing industry-standard measurements for dexterity that previous benchmarks have not covered.

To promote its innovations and reach a global audience, RLWRLD has been hosting a series of events called Dexterity Night. The inaugural event held in San Francisco featured Amit Goel from NVIDIA, who praised RLWRLD as a pivotal partner within NVIDIA's developing physical AI ecosystem. Following the event in Japan, RLWRLD is set to hold another Dexterity Night in Seoul on June 10.

About RLWRLD

Founded in 2024 and headquartered in the U.S., RLWRLD is a physical AI company dedicated to developing robotic models that impart human-like dexterity and cognitive abilities. With operations in South Korea and Japan, RLWRLD operates a unique system for collecting and training high-precision multimodal industrial data. This enables robots to comprehend and respond to real-world industrial environments effectively. The company is actively undertaking pilot projects with key industry partners, aiming to become a global leader in industrial robot AI.