Kairos: A Native World Model Stack for Physical AI

2026-06-15

AI Summary

Existing world models struggle to meet the demands of embodied intelligence for efficiency, robustness, and long-horizon state maintenance. This work proposes Kairos—the first end-to-end native world model stack—designed to progressively pretrain through a curriculum spanning diverse embodied data. Kairos employs a unified architecture that seamlessly integrates perception, generation, and prediction, and introduces a hybrid linear temporal attention mechanism combining sliding windows, dilated windows, and gated linear attention, with theoretical guarantees on bounded error accumulation. Co-designed with deployment-aware considerations, Kairos achieves state-of-the-art performance across embodied modeling, long-horizon tasks, and policy benchmarks, while substantially improving the trade-off between efficiency and capability, enabling real-time observation-to-action loops on consumer-grade hardware.

Abstract

World models are transitioning from passive visual generators to foundational, operational infrastructure for Physical AI: they must natively acquire world knowledge from heterogeneous experience, maintain persistent states over long horizons, and execute efficiently within real deployment constraints. We introduce Kairos, a native world model stack designed around these requirements. (1) Kairos learns the world by pioneering a Native Pre-training Paradigm governed by a Cross-Embodiment Data Curriculum, which organizes open-world videos, human behavioral data, and robot interactions into a progressive developmental pathway. (2) Kairos maintains the world by unified world understanding, generation, and prediction within a Native Unified Architecture equipped with Hybrid Linear Temporal Attention, where sliding-window attention captures local dynamics, dilated sliding windows capture mid-range dependencies, and gated linear attention maintains persistent global memory. We establish formal theoretical bounds demonstrating that this temporal factorization strictly limits error accumulation, mathematically guaranteeing state propagation across extended horizons. (3) Kairos runs the world by incorporating a Deployment-Aware System Co-Design to support low-latency rollout generation on server and consumer-grade hardware for real-world observation-action-feedback loops. Experiments on embodied world-model, long-horizon, and action-policy benchmarks show that Kairos achieves top level performance while offering a strong efficiency-capability trade-off. Together, these results position Kairos as a cohesive operational foundation for future self-evolving physical intelligence.

Problem

• world models
• Physical AI
• long-horizon state maintenance
• heterogeneous experience
• deployment constraints

Innovation

• Native World Model
• Cross-Embodiment Data Curriculum
• Hybrid Linear Temporal Attention
• Deployment-Aware Co-Design
• Error Accumulation Bound