At WAIC 2026, Tencent Robotics X Laboratory open-sourced three embodied foundation models: Hy-Embodied-VLM-1.0 for spatial and scene understanding, Hy-Embodied-RxBrain-1.0 for cognitive planning with visual state imagination, and Hy-Embodied-VLA-0.5 for converting high-level goals into continuous motor commands. Chief Scientist Dr. Zhang Zhengyou presented the architecture as a solution to a fundamental problem: different types of robotic intelligence must operate at different frequencies because the physical world operates across multiple time scales. The insight stemmed from a failed experiment porting robot capabilities into OpenClaw, which caused tens of seconds of delay unacceptable for physical robots where even 2-3 seconds of cognitive latency is already problematic. The three-layer architecture addresses this through frequency separation, with Hy-Embodied-VLA operating at the highest frequency for real-time physical interaction. Tencent disclosed that Hy-Embodied-VLA has entered production testing at a daily chemical factory, handling high-mix, low-batch, SKU-intensive assembly lines with single-piece cycle time under 6 seconds, and can adapt to new SKUs with only 8 hours of data collection and fine-tuning. Dr. Zhang emphasized that a demo scoring 80-90 points without actual deployment is effectively zero value.