InternVLA-N1 adopts a compositional architecture featuring a dual-system design that synergistically combines high-level instruction interpretation with low-level action execution. Specifically, our system integrates:

• System 2: A vision-language model (VLM)-based planning module that interprets navigation instructions to predict mid-term waypoint goals through image-grounded reasoning. By predicting pixel coordinates in the image space, it effectively connects instruction understanding with spatial reasoning, enabling long-horizon navigation instruction following.
• System 1: A multi-modal goal-conditioned diffusion policy guided by latent plan or supported explicit goals, which generates executable short-horizon trajectories conditioned on current observations and the asynchronous latent features from System 2. It enables robust, real-time control and local decision-making in complex environments.

To fully unlock open-world generalization and asynchronous inference capabilities in the dual-system architecture, we design a curriculum training scheme. Initially, each system is trained separately to acquire basic navigation skills using explicit goals in a synchronized setting. Then, a joint fine-tuning phase is introduced. In this phase, we incorporate learnable tokens into System 2 as implicit midterm goals to reduce the ambiguity of pixel-based targets.

For the navigation task, most real-world datasets are constrained by scene diversity and scale. Meanwhile, Internet video datasets suffer from imprecise localization and mapping information, which limits their feasibility as reliable navigation datasets for trajectory prediction. In contrast, we propose efficient pipelines for generating navigation datasets in simulation, aiming to facilitate scalable training.