Kairos: A Regret-Aware Native World-Action Model Stack for Physical AI

We introduce $\mathbf{\text{Kairos}}$, a regret-aware native world-action model stack for Physical AI. Kairos is motivated by the view that a physical world model should not aim to fully simulate all future pixels, but should learn and maintain the information most relevant to embodiment control: object state, spatial relations, contact conditions, task progress, action consequences, failure boundaries, and deployment uncertainty. Kairos establishes three model-side prerequisites toward this goal. First, it $\mathbf{\text{learns}}$ control-relevant information through a $\mathbf{\text{Cross-Embodiment}}\mathbf{\text{Data}}\mathbf{\text{Curriculum}}$, which organizes open-world videos, human behavioral data, and robot interactions into an intervention-strength progression from passive physical observation to intentional behavior and embodied action grounding. Second, it $\mathbf{\text{maintains}}$ control-sufficient states through a unified $\mathbf{\text{understanding,}}\mathbf{\text{generation,}}\mathbf{\text{and}}\mathbf{\text{prediction}}\mathbf{\text{architecture}}$ equipped with $\mathbf{\text{Hybrid}}\mathbf{\text{Linear}}\mathbf{\text{Temporal}}\mathbf{\text{Attention}}$, where local, mid-range, and global temporal pathways support multi-timescale state maintenance under efficient inference. Third, it $\mathbf{\text{deploys}}$ these states through a $\mathbf{\text{Deployment-Aware}}\mathbf{\text{System}}\mathbf{\text{Co-Design}}$, treating latency, memory footprint, and hardware compatibility as first-order constraints for future observation, action, and feedback loops. Experiments on embodied world-model benchmarks, world-action benchmarks, long-horizon generation, and inference-efficiency evaluation show that Kairos achieves superior performance while offering a favorable efficiency to capability trade-off.