Putting Dexterous Robots to Work: How RLWRLD Builds Physical AI with AWS

This article describes how RLWRLD, a Physical AI company, uses AWS infrastructure to train RLDX-1, a robotics foundation model designed for dexterous manipulation tasks with five-finger hands.

• RLDX-1 is an 8.1B-parameter open-source model that fuses vision-language understanding with proprioception, tactile, and torque sensing
• Trained on hundreds of terabytes of real-world robot data collected from actual factory and service floors using AWS EC2 p5e/p5en instances with NVIDIA H200 GPUs
• Uses AWS ParallelCluster for distributed training orchestration and Amazon FSx for Lustre for high-speed concurrent data access
• Achieves state-of-the-art results on RoboCasa, LIBERO, and SIMPLER benchmarks using approximately 20% of training compute of comparable models
• Deployed at LOTTE HOTEL & RESORT in Seoul to capture real service-skill data for tasks like linen folding and glassware handling

RLDX-1 enables robots to perform fine motor tasks requiring human-like dexterity in real industrial environments, addressing the last mile of automation where most physical work still demands skills technology cannot yet reliably replicate.