NavGPT-3: Harnessing Context in a Hierarchical Navigation Runtime

2026-10-07

AI Summary

This work addresses the embodied navigation bottleneck arising from the difficulty of coordinating large-model reasoning with low-latency physical control by introducing a pioneering operating system-inspired hierarchical runtime architecture. Through a thread-scheduling mechanism, the proposed framework integrates long-horizon language model reasoning with vision-intensive action control while supporting real-time interrupt responses. By combining long-horizon reinforcement learning, an 8B Vision-Language-Action (VLA) policy, and scene-change-aware dynamic allocation of Codec visual tokens, the system reduces reaction latency to the sub-second level. Extensive evaluations demonstrate state-of-the-art performance on R2R-CE with a success rate (SR) of 81.51. Notably, on RxR-CE, the approach achieves an SR of 90.43, surpassing human-level performance (90.4) for the first time. These results validate the effectiveness of the proposed architecture in enabling efficient, autonomous embodied navigation.

Abstract

Language models trained with long-horizon agentic reinforcement learning can generalize knowledge through reasoning, express precise actions, and pursue goals over many steps, raising the ceiling on what an embodied agent can understand and decide. Physical interaction, however, remains the domain of action policies, which provide dense, low-latency control. We present NavGPT-3, a harness that connects the two models, with an OS-like runtime built above it: reasoning, acting, and monitoring run as threads with their own context, tools, and permissions, while the runtime schedules them and decides which thread controls the robot's motion, so that the robot can react to sudden real-world events through interruption and thread switching. Beneath it, our action policy NavGPT VLA, trained on 19.28M examples, allocates visual tokens using codec allocation, in proportion to scene change; its 8B model alone reaches 74.51 SR on R2R-CE and leads RxR-CE with 78.19 SR. With the complete harness, NavGPT-3 sets the state of the art on R2R-CE (81.51 SR) and, for the first time, brings an autonomous agent to human level: on RxR-CE it matches human followers in success (90.43 vs. 90.4 SR) and path fidelity (78.47 vs. 77.7 nDTW) at 1 min 22 s per episode, versus roughly 3 min for a human. We comprehensively ablate the harness design and the interaction between the two models, showing how tools and the action policy shape the path from language-model reasoning to physical control: when NavGPT VLA executes the route, the reasoning loop shortens and the system's minimum reaction time falls from 3-19 s per language-model decision to 0.5-1 s per action-policy step (1-2 Hz). These results show that designing this embodied interface is central to connecting frontier language-model intelligence with low-level physical control. We will release all models, code, and evaluation records.

Problem

• Embodied Navigation
• Language Models
• Action Policies
• Hierarchical Runtime
• Vision-Language-Action

Innovation

• Hierarchical Navigation Runtime
• Vision-Language-Action Model
• Codec Token Allocation
• Embodied Agent
• Thread Scheduling