ONEROBOTICS (06600) has officially launched its proprietary world action model, OneModel 1.7 FrontoStria-RL. Designed for real-world deployment in home and service robotics, this model serves as the core foundational platform for ONEROBOTICS' initiative to scale the implementation of embodied intelligence. On the LIBERO benchmark for embodied intelligence, OneModel 1.7 achieved an average success rate of 99%, outperforming mainstream public models such as π0.5, GR00T-N1.5, and OpenVLA-OFT. In actual robot deployments, it attained a 99% success rate for daily operational tasks and a 97% success rate for high-precision tasks.

The model is built on ONEROBOTICS' self-designed RL-Latent World Action Model (RL-LWAM) architecture, which comprises three key modules. The World Model handles cross-scenario generalization, the Understand Expert manages task comprehension and skill scheduling, and the Action Expert ensures precise execution. These components are implicitly connected via a Predictive Policy Latent mechanism, eliminating the need for explicit intermediate image or coordinate transmission.

Furthermore, a reinforcement learning loop integrated with a success memory mechanism continuously feeds real-world deployment feedback back into the model, enabling its capabilities to grow with the scale of deployment. The prerequisite for embodied robots to enter real-life scenarios extends beyond task understanding to the reliable completion of tasks. OneModel 1.7 FrontoStria-RL integrates generalized understanding, action success rates, and real feedback learning into a unified world action model system. It covers a broad spectrum of scenarios, from routine household operations to high-precision, high-dynamic interactions. This is not merely a single-point demonstration but a model platform validated by test results and capable of scalable delivery.