GPT-6-Astra Lights Up Embodied Navigation: Evaluation in Zero-Shot Vision-and-Language Navigation in Continuous Environments

We investigate whether GPT-6-Astra, a general-purpose foundation model, can navigate unfamiliar environments using its own perception, reasoning, and decision-making capabilities. Our evaluation focues on zero-shot vision-and-language navigation in continuous environments (VLN-CE) through a minimal interface in the Codex harness, aiming to unleash GPT-6-Astra’s full potential for navigation. Using monocular RGB, GPT-6-Astra decides when to observe, how to move, and when to stop, without navigation-specific fine-tuning, a trained waypoint predictor, or a pre-built scene map. Our evaluation yields four main findings. First, $\mathit{\text{GPT-6-Astra}}\mathit{\text{achieves}}\mathit{\text{strong}}\mathit{\text{zero-shot}}\mathit{\text{navigation}}\mathit{\text{performance}}\mathit{\text{using}}\mathit{\text{only}}\mathit{\text{monocular}}\mathit{\text{RGB}}\mathit{\text{observations}}$. On the common-adopted zero-shot R2R-CE benchmark, ultra reasoning achieves a success rate of $\mathit{\text{79.0\%}}$, exceeding the strongest reported zero-shot and supervised success rates by $\mathit{\text{13.0}}$ and $\mathit{\text{6.9}}$ percentage points, respectively. Second, $\mathit{\text{GPT-6-Astra}}\mathit{\text{advances}}\mathit{\text{multi-stage}}\mathit{\text{language}}\mathit{\text{instructions}}\mathit{\text{into}}\mathit{\text{coherent,}}\mathit{\text{adaptive}}\mathit{\text{navigation}}$ by grounding spatial relations, tracking task progress, and revising its actions. Third, $\mathit{\text{reliable}}\mathit{\text{route}}\mathit{\text{execution}}\mathit{\text{and}}\mathit{\text{goal}}\mathit{\text{verification}}\mathit{\text{remain}}\mathit{\text{challenging,}}\mathit{\text{even}}\mathit{\text{with}}\mathit{\text{ultra}}\mathit{\text{reasoning}}$. Plausible local landmark matches do not consistently lead to correct task completion. Fourth, $\mathit{\text{these}}\mathit{\text{capabilities}}\mathit{\text{motivate}}\mathit{\text{rethinking}}\mathit{\text{the}}\mathit{\text{role}}\mathit{\text{of}}\mathit{\text{embodied}}\mathit{\text{learning}}$. Future VLN research should build on foundation models to advance generalizable and reliable embodied intelligence.