RLWRLD Inc. is a robotics AI company that builds foundation models for robotic dexterity, enabling robots to perform complex, human-hand-level manipulation tasks in real-world industrial settings. Their flagship product, RLDX-1, is a dexterity-first foundation model that integrates vision, force sensing, and memory to support deployment across single-arm, dual-arm, and humanoid robot embodiments.
“The paper's central contribution is a framework (H2RB) that lets you adapt a pretrained robot policy to new tasks and objects using only human video demonstrations — no paired human-robot data, no target robot teleoperation for the new tasks.”
Source→“When H2RB is combined with large-scale heterogeneous training across multiple embodiments (14,548 episodes, 5M+ frames from EgoDex, in-house human data, OpenArm, and ALLEX robot data), performance improves further — skill transfer rises from 62.7 to 81.7, and object transfer from 51.8 to 57.2”
Source→AI-extracted from podcast / newsletter / paper summaries. May contain errors.