Kairos: A Native World Model Stack for Physical AI

Kairos is introduced as a comprehensive world model stack for Physical AI, featuring a Native Pre-training Paradigm that utilizes a Cross-Embodiment Data Curriculum to learn from diverse data sources. It employs a Native Unified Architecture with Hybrid Linear Temporal Attention to manage world understanding, generation, and prediction, ensuring minimal error accumulation over extended time horizons. Kairos is designed for efficient deployment on various hardware, achieving superior performance in embodied world-model and action-policy benchmarks, making it a significant advancement for practitioners developing self-evolving physical intelligence systems.