RLDX-1: Open-Source Robotics Foundation Model Achieves 86.8% on Humanoid Tasks — Doubles π₀.₅ and GR00T N1.6

arXiv·medium signal

RLWRLD released RLDX-1, an 8.1B-parameter open-source robotics foundation model using a Multi-Stream Action Transformer (MSAT) that processes vision, language, torque, tactile feedback, and working memory via joint attention. On ALLEX humanoid benchmarks, RLDX-1 hits 86.8% success rate vs ~40% for π₀.₅ and GR00T N1.6. Three checkpoints released on HuggingFace under open license. 67 co-authors.