NavHarness: Towards Lifelong Embodied Navigation
NavHarness lifts lifelong embodied navigation, gaining up to 22.6 s-SR points on GOAT-Bench.
NavHarness is a training-free embodied navigation harness that treats maps, task records, and house knowledge as part of each multi-round reasoning loop. It checks stored experience against observations, records corrections, and carries that state into new conversations for later tasks or recovery. On GOAT-Bench it improves success-rate over context-only sessions by 18.6 points with Astra and 22.6 with Opus 5. With SLAM-estimated poses, GPT-6 Astra reaches 83.7 s-SR and 36.9 e-SR on GOAT-Bench and 85.9 s-SR on IR2R-CE.
- Training-free harness checks maps and task records against new observations.
- GOAT-Bench s-SR rises 18.6 points with Astra and 22.6 with Opus 5.
- GPT-6 Astra reaches 83.7 s-SR and 36.9 e-SR on GOAT-Bench.
- SLAM-pose setup scores 85.9 s-SR on IR2R-CE.
- Structured recovery handovers beat length-matched summaries for reuse.
- Products
- NavHarness
- AI models
- AstraOpus 5GPT-6 Astra
From the original article
1 min read at huggingface.co
Frontier models can now perform well on individual embodied navigation tasks through multi-round multimodal reasoning with simple tools. Across successive tasks, however, an agent must also rely on an evolving map and earlier search records, both of which may be incomplete or conflict with new observations.
By Xunyi Zhao, Jian Zhou, Sihao Lin · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.