Real-World (RLWRLD) Launches Robot Hand Analysis Platform "All Hands Up!" to Provide Common Benchmarks for the Robot Hand Ecosystem

Real-World (RLWRLD), a Physical AI innovator led by its proprietary Robotics Foundation Model (RFM) "RLDX-1 (RealDex)", has launched "All Hands Up!" (www.allhandsup.org). This website provides a technical report and visualization features that organize the practical limitations and trade-offs of dexterous robot hands based on operational data from commercially available models. Real-World is headquartered in the US with CEO Ryu Jun-hee, and its Japanese subsidiary is located in Chiyoda-ku, Tokyo, with Representative Lee Hoon.

"All Hands Up!" is a platform that analyzes and discloses the actual operational performance and design trade-offs of robot hands, which are difficult to grasp from manufacturer specification sheets alone. We developed and launched this site based on accumulated operational data to answer the recurring question in research and industry: "Which robot hand functions effectively in real-world environments?"

Structural "Trade-offs" in Robot Hand Design

Robot hands are considered core elements of Physical AI. However, due to structural trade-offs between size, gripping force, and backdrivability, developing products that satisfy all performance aspects simultaneously is currently difficult.

Relationship between Size and Gripping Force: When the size of a robot hand is reduced, the internal drive motor also becomes smaller, leading to a decrease in gripping force.

Relationship between Gripping Force and Backdrivability: Increasing the gear ratio (the ratio of gears meshing with the motor) to enhance force strengthens the gripping force, but it reduces "backdrivability," the characteristic of flexibly responding to external forces and impacts.

Thus, current commercial products have limitations and trade-offs depending on their design objectives, as improving one aspect often compromises another.

Presenting Objective Indicators with Proprietary Benchmark "DexBench"

To more accurately evaluate actual task performance, Real-World has organized key design variables that affect operational efficiency in the field:

Thumb range of motion (Kapandji Scale)

Independent drive capability of distal interphalangeal (DIP) joints

Minimum graspable diameter (the smallest object diameter the robot hand can grip)

Friction characteristics of the hand's outer material

In addition to these, using our proprietary benchmark "DexBench," we quantitatively analyze the characteristics and limitations of each robot hand based on 18 real-world manipulation tasks.

"Hardware Dualization" Strategy for the Reality of No Perfect Hand

Acknowledging the current reality that a perfect robot hand does not yet exist, we propose a practical strategy of dualizing hardware for operation according to its intended use:

For Field Deployment (Type 1): A practical structure prioritizing lightweight and high durability, considering actual industrial environments.

For Learning Data Collection (Type 2): A structure with high backdrivability and precision, enabling fine manipulation and data acquisition for AI learning.

We believe that mutually complementary utilization of these two types is the optimal approach in current robot hand development.

"URDF-Based Visualization Function" Verifiable in Web Browsers

"All Hands Up!" provides interactive visualization information based on URDF (a standard format for robot description) for robot hands evaluated by Real-World.

Users can verify whether their desired grasp shape can be achieved by manipulating the joints of multiple robot hands with mouse operations in a web browser, without the need for expensive specialized software or individual development environments. We also provide URDF data that can be directly used for actual robot simulations and development, in addition to comparisons of key specifications for each product (currently includes data for over 10 types of multi-joint robot hands).

Comment from Ryu Jun-hee, CEO of Real-World

"'All Hands Up!' is not just a product comparison website; it is an open platform for sharing robot hand operational data with the entire industry. We will not stop at a single release but will continuously accumulate the latest robot hand verification data through quarterly content updates. This will allow manufacturers to validate their designs, and researchers and industrial partners to establish clear criteria for robot hand adoption. By providing common benchmarks, we aim to contribute to the development of the robot hand development ecosystem."

About Real-World (RLWRLD)

Real-World is a Physical AI company developing Robotics Foundation Models (RFMs) that implement human-level hand dexterity and cognitive abilities in machines. Founded in 2024, the company has offices in the US, South Korea, and Japan. Based on its unique system for collecting and learning high-precision 4D+ multimodal industrial data, it leads the technology for robots in industrial environments to perceive, understand, and act like humans. Currently, the company is proceeding with commercialization based on pilot projects with major companies in South Korea and Japan, aiming to become a global leader in the industrial robot AI field.

Founded: 2024

Locations: USA, South Korea, Japan

Business Activities: Development of Robotics Foundation Models (RFM), provision of Physical AI solutions

Official Website: www.allhandsup.org (Special site for this platform)

Official HP: https://www.rlwrld.ai/jp