OneModel 1.7 FrontoStria-RL

Overview

OneModel 1.7 FrontoStria-RL is a world action model (WAM) developed by OneRobotics for real-world household and service robot deployment. It uses a self-designed RL-Latent World Action Model (RL-LWAM) architecture that connects high-level world understanding with action execution via Predictive Policy Latent. The system integrates a World Model for cross-scenario generalization, an Understand Expert for task decomposition and skill scheduling, and an Action Expert for precise execution, all within a reinforcement learning closed loop with success memory for continuous improvement. It achieves 99% average task success rate on the LIBERO benchmark, ~99% on daily manipulation tasks, ~97% on high-precision tasks, and 91.2% receive success in table-tennis rallying with a human. Validated on LIBERO, LIBERO-plus, SimplerEnv benchmarks and real-robot tests including laundry handling, clothes folding, dishwasher operation, test-tube handling, cup stacking, and table-tennis rally.