{"id":"166899d74667ced3b5e38c1d51aafb8466b8e310564d1bf8f77d6055212f08f5","title":"What Should a Navigator Remember? Spatiotemporal Memory for Zero-Shot Vision-Language Navigation — arXiv cs.AI","argument":"The authors study what an 8B-scale navigation system should retain when following language instructions without navigation-specific training. Their system organizes information into task, temporal, and spatial memories: instruction progress, observation history anchored to poses, and accumulated geometry. An 8B observer summarizes observations; an 8B reasoner uses the memories to choose waypoint intents or request more views, while a geometric controller executes selected targets. The authors distinguish retaining evidence, making it accessible to the reasoner, and acquiring further observations. On R2R-CE val-unseen, the complete system achieved the highest result among the evaluated Qwen3-based zero-shot baselines. Removing temporal memory reduced SR and sharply increased average path length from 16.9 to 39.1m. Removing the spatial-map and planning component also reduced SR. In real-world corridor and office experiments with a Unitree Go2, the zero-shot method achieved higher success rates than a post-trained baseline. The authors say these results suggest better transfer to the evaluated real-world settings. They also use paired instruction interventions in LAW2 to examine instruction-following behavior beyond endpoint success.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The supplied evidence is the abstract, not the full paper; it does not provide the detailed evaluation setup or results underlying the reported comparisons. The real-world transfer result is limited in the abstract to the evaluated corridor and office environments. The abstract reports higher success rates than the post-trained baseline but gives no numerical rates.","speaker":{"name":"Peng, Yuhang, Wang, Zhikai, Wang, Ziyi, Zhong, Jiaru, Zheng, Xiaoji, Huang, Jingzhi, Wang, Yi, Gong, Jiangtao","roleAtSource":"Article author"},"source":{"title":"What Should a Navigator Remember? Spatiotemporal Memory for Zero-Shot Vision-Language Navigation","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2610.38318","publishedAt":"2026-10-28","dateBasis":"Date basis not provided."},"evidence":[{"text":"The system combines task, temporal, and spatial memories for later navigation decisions.","excerpt":"Our agent combines instruction progress, hierarchically compressed observation history anchored to poses, and accumulated geometry in complementary task, temporal, and spatial memories.","url":"https://arxiv.org/abs/2610.38318","relationship":"Source-stated"},{"text":"The system uses an observer to summarize observations and a reasoner to choose targets or request more views, with a controller executing targets.","excerpt":"An 8B observer summarizes observations, and an 8B reasoner uses the memories to select waypoint intents or request additional views, a geometric controller executes the selected targets.","url":"https://arxiv.org/abs/2610.38318","relationship":"Source-stated"},{"text":"On R2R-CE val-unseen, the complete system ranked highest among the evaluated Qwen3-based zero-shot baselines.","excerpt":"On R2R-CE val-unseen, the complete system achieves the highest among the evaluated Qwen3-based zero-shot baselines.","url":"https://arxiv.org/abs/2610.38318","relationship":"Source-stated"},{"text":"Removing temporal memory reduced SR and increased average path length; removing the spatial-map and planning component also reduced SR.","excerpt":"Removing temporal memory reduces SR and sharply increases average path length from 16.9 to 39.1m, removing the spatial-map and planning component reduces SR as well.","url":"https://arxiv.org/abs/2610.38318","relationship":"Source-stated"},{"text":"In evaluated corridor and office experiments, the zero-shot method had higher success rates than the post-trained baseline.","excerpt":"Real-world experiments on a Unitree Go2 show that our zero-shot method achieves higher success rates than the post-trained baseline in both corridor and office environments","url":"https://arxiv.org/abs/2610.38318","relationship":"Source-stated"},{"text":"The authors say their real-world results suggest better transfer to the settings they evaluated.","excerpt":"suggesting better transfer to the evaluated real-world settings.","url":"https://arxiv.org/abs/2610.38318","relationship":"Source-stated"},{"text":"The authors use paired instruction interventions to examine behavioral responses beyond endpoint success.","excerpt":"As a complementary evaluation, LAW2 uses paired instruction interventions to examine behavioral responses, instruction following ability beyond endpoint success.","url":"https://arxiv.org/abs/2610.38318","relationship":"Source-stated"},{"text":"The authors conclude that structured memory preserving task progress, observation history, and geometry supports zero-shot navigation with smaller open models.","excerpt":"These results support preserving task progress, observation history, and geometry in structured memory for zero-shot navigation with smaller open models.","url":"https://arxiv.org/abs/2610.38318","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2610.38318","checkedAt":"2026-11-11T10:42:10.072Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"What evidence from the evaluated corridor and office experiments underlies the reported higher success rates and transfer claim?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"What Should a Navigator Remember? Spatiotemporal Memory for Zero-Shot Vision-Language Navigation — arXiv cs.AI","summary":"The authors study what an 8B-scale navigation system should retain when following language instructions without navigation-specific training. Their system organizes information into task, temporal, and spatial memories: instruction progress, observation history anchored to poses, and accumulated geometry. An 8B observer summarizes observations; an 8B reasoner uses the memories to choose waypoint intents or request more views, while a geometric controller executes selected targets. The authors distinguish retaining evidence, making it accessible to the reasoner, and acquiring further observations. On R2R-CE val-unseen, the complete system achieved the highest result among the evaluated Qwen3-based zero-shot baselines. Removing temporal memory reduced SR and sharply increased average path length from 16.9 to 39.1m. Removing the spatial-map and planning component also reduced SR. In real-world corridor and office experiments with a Unitree Go2, the zero-shot method achieved higher success rates than a post-trained baseline. The authors say these results suggest better transfer to the evaluated real-world settings. They also use paired instruction interventions in LAW2 to examine instruction-following behavior beyond endpoint success.","question":"What evidence from the evaluated corridor and office experiments underlies the reported higher success rates and transfer claim?","claims":[{"text":"The system combines task, temporal, and spatial memories for later navigation decisions.","excerpt":"Our agent combines instruction progress, hierarchically compressed observation history anchored to poses, and accumulated geometry in complementary task, temporal, and spatial memories.","support":"source-stated"},{"text":"The system uses an observer to summarize observations and a reasoner to choose targets or request more views, with a controller executing targets.","excerpt":"An 8B observer summarizes observations, and an 8B reasoner uses the memories to select waypoint intents or request additional views, a geometric controller executes the selected targets.","support":"source-stated"},{"text":"On R2R-CE val-unseen, the complete system ranked highest among the evaluated Qwen3-based zero-shot baselines.","excerpt":"On R2R-CE val-unseen, the complete system achieves the highest among the evaluated Qwen3-based zero-shot baselines.","support":"source-stated"},{"text":"Removing temporal memory reduced SR and increased average path length; removing the spatial-map and planning component also reduced SR.","excerpt":"Removing temporal memory reduces SR and sharply increases average path length from 16.9 to 39.1m, removing the spatial-map and planning component reduces SR as well.","support":"source-stated"},{"text":"In evaluated corridor and office experiments, the zero-shot method had higher success rates than the post-trained baseline.","excerpt":"Real-world experiments on a Unitree Go2 show that our zero-shot method achieves higher success rates than the post-trained baseline in both corridor and office environments","support":"source-stated"},{"text":"The authors say their real-world results suggest better transfer to the settings they evaluated.","excerpt":"suggesting better transfer to the evaluated real-world settings.","support":"source-stated"},{"text":"The authors use paired instruction interventions to examine behavioral responses beyond endpoint success.","excerpt":"As a complementary evaluation, LAW2 uses paired instruction interventions to examine behavioral responses, instruction following ability beyond endpoint success.","support":"source-stated"},{"text":"The authors conclude that structured memory preserving task progress, observation history, and geometry supports zero-shot navigation with smaller open models.","excerpt":"These results support preserving task progress, observation history, and geometry in structured memory for zero-shot navigation with smaller open models.","support":"source-stated"}],"limitations":["The supplied evidence is the abstract, not the full paper; it does not provide the detailed evaluation setup or results underlying the reported comparisons.","The real-world transfer result is limited in the abstract to the evaluated corridor and office environments.","The abstract reports higher success rates than the post-trained baseline but gives no numerical rates."],"mediaId":"a391043ca83168f1c2a6d22029b891db2181efa62fdd497331ca3739fca2aba8","correctionReason":null,"id":"166899d74667ced3b5e38c1d51aafb8466b8e310564d1bf8f77d6055212f08f5","publicationId":"7f4a3397-c633-436b-8496-c65393c4cd32","version":1,"sourceUrl":"https://arxiv.org/abs/2610.38318","sourceTitle":"What Should a Navigator Remember? Spatiotemporal Memory for Zero-Shot Vision-Language Navigation","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-10-28","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-10-30T06:40:04.841Z","createdAt":"2026-11-11T10:11:06.372Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"a391043ca83168f1c2a6d22029b891db2181efa62fdd497331ca3739fca2aba8","kind":"article","url":"https://arxiv.org/abs/2610.38318","sourceUrl":"https://arxiv.org/abs/2610.38318","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-10-31T04:32:22.403Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2610.38318","checkedAt":"2026-11-11T10:42:10.072Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"a391043ca83168f1c2a6d22029b891db2181efa62fdd497331ca3739fca2aba8","confidence":0.6900000000000001,"provider":"typesafe","model":"jev-1.13.0","inputHash":"639a6eff9c3c577ac1df3d0dceded899440e25ac6402b912eaceed23167693e7","decidedAt":"2026-10-31T04:32:23.514Z","redecided":{"from":"stream-admission-v5","under":"stream-admission-v5+confidence-0.5"}},"provenance":{"textHash":"e091eb22c08e01ae1a86700997737cd1420e03b19aa17f53f4cc4f2f5d825862","observationId":"a608345b7327b4e76ee7c2cb0636876ebf842ad1e957fc6b8fb913b8db19845a","retrievedAt":"2026-10-31T04:32:22.231Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"0003fc36-87ee-42d7-8611-8023837d17f7"},"reviewer":{"model":"gpt-6-luna","runId":"d7a7760a-8e6d-42fb-8970-afc84399b92c"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-11T10:11:06.372Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Peng, Yuhang","url":null,"kind":"unknown"},{"name":"Wang, Zhikai","url":null,"kind":"unknown"},{"name":"Wang, Ziyi","url":null,"kind":"unknown"},{"name":"Zhong, Jiaru","url":null,"kind":"unknown"},{"name":"Zheng, Xiaoji","url":null,"kind":"unknown"},{"name":"Huang, Jingzhi","url":null,"kind":"unknown"},{"name":"Wang, Yi","url":null,"kind":"unknown"},{"name":"Gong, Jiangtao","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Peng, Yuhang","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Peng, Yuhang","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.38318","locator":"head/meta[name=author|citation_author]"},"assertionId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-10-31T04:32:22.231Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a","sourceUrl":"https://arxiv.org/abs/2610.38318","contentSha256":"2b26b73da229541cc3dc77ca355d9e7e8f77e838ec9c48be769c1209d7b2820c","textHash":"e091eb22c08e01ae1a86700997737cd1420e03b19aa17f53f4cc4f2f5d825862","observedAt":"2026-10-31T04:32:22.231Z"}},{"actor":{"name":"Wang, Zhikai","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Wang, Zhikai","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.38318","locator":"head/meta[name=author|citation_author]"},"assertionId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-10-31T04:32:22.231Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a","sourceUrl":"https://arxiv.org/abs/2610.38318","contentSha256":"2b26b73da229541cc3dc77ca355d9e7e8f77e838ec9c48be769c1209d7b2820c","textHash":"e091eb22c08e01ae1a86700997737cd1420e03b19aa17f53f4cc4f2f5d825862","observedAt":"2026-10-31T04:32:22.231Z"}},{"actor":{"name":"Wang, Ziyi","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Wang, Ziyi","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.38318","locator":"head/meta[name=author|citation_author]"},"assertionId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-10-31T04:32:22.231Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a","sourceUrl":"https://arxiv.org/abs/2610.38318","contentSha256":"2b26b73da229541cc3dc77ca355d9e7e8f77e838ec9c48be769c1209d7b2820c","textHash":"e091eb22c08e01ae1a86700997737cd1420e03b19aa17f53f4cc4f2f5d825862","observedAt":"2026-10-31T04:32:22.231Z"}},{"actor":{"name":"Zhong, Jiaru","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Zhong, Jiaru","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.38318","locator":"head/meta[name=author|citation_author]"},"assertionId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-10-31T04:32:22.231Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a","sourceUrl":"https://arxiv.org/abs/2610.38318","contentSha256":"2b26b73da229541cc3dc77ca355d9e7e8f77e838ec9c48be769c1209d7b2820c","textHash":"e091eb22c08e01ae1a86700997737cd1420e03b19aa17f53f4cc4f2f5d825862","observedAt":"2026-10-31T04:32:22.231Z"}},{"actor":{"name":"Zheng, Xiaoji","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Zheng, Xiaoji","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.38318","locator":"head/meta[name=author|citation_author]"},"assertionId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-10-31T04:32:22.231Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a","sourceUrl":"https://arxiv.org/abs/2610.38318","contentSha256":"2b26b73da229541cc3dc77ca355d9e7e8f77e838ec9c48be769c1209d7b2820c","textHash":"e091eb22c08e01ae1a86700997737cd1420e03b19aa17f53f4cc4f2f5d825862","observedAt":"2026-10-31T04:32:22.231Z"}},{"actor":{"name":"Huang, Jingzhi","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Huang, Jingzhi","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.38318","locator":"head/meta[name=author|citation_author]"},"assertionId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a:participation:5","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-10-31T04:32:22.231Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a","sourceUrl":"https://arxiv.org/abs/2610.38318","contentSha256":"2b26b73da229541cc3dc77ca355d9e7e8f77e838ec9c48be769c1209d7b2820c","textHash":"e091eb22c08e01ae1a86700997737cd1420e03b19aa17f53f4cc4f2f5d825862","observedAt":"2026-10-31T04:32:22.231Z"}},{"actor":{"name":"Wang, Yi","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Wang, Yi","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.38318","locator":"head/meta[name=author|citation_author]"},"assertionId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a:participation:6","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-10-31T04:32:22.231Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a","sourceUrl":"https://arxiv.org/abs/2610.38318","contentSha256":"2b26b73da229541cc3dc77ca355d9e7e8f77e838ec9c48be769c1209d7b2820c","textHash":"e091eb22c08e01ae1a86700997737cd1420e03b19aa17f53f4cc4f2f5d825862","observedAt":"2026-10-31T04:32:22.231Z"}},{"actor":{"name":"Gong, Jiangtao","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Gong, Jiangtao","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.38318","locator":"head/meta[name=author|citation_author]"},"assertionId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a:participation:7","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-10-31T04:32:22.231Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a","sourceUrl":"https://arxiv.org/abs/2610.38318","contentSha256":"2b26b73da229541cc3dc77ca355d9e7e8f77e838ec9c48be769c1209d7b2820c","textHash":"e091eb22c08e01ae1a86700997737cd1420e03b19aa17f53f4cc4f2f5d825862","observedAt":"2026-10-31T04:32:22.231Z"}}],"participationCoverage":{"total":8,"returned":8,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2610.38318","snapshotId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a","contentSha256":"2b26b73da229541cc3dc77ca355d9e7e8f77e838ec9c48be769c1209d7b2820c","textHash":"e091eb22c08e01ae1a86700997737cd1420e03b19aa17f53f4cc4f2f5d825862","observedAt":"2026-10-31T04:32:22.231Z","basis":"publisher-page-metadata","sourceLinks":47,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/166899d74667ced3b5e38c1d51aafb8466b8e310564d1bf8f77d6055212f08f5/source"}},"canonicalUrl":"https://escapemeridian.com/readings/what-should-a-navigator-remember-spatiotemporal-memory-for-zero-166899d7","formats":{"html":"/readings/what-should-a-navigator-remember-spatiotemporal-memory-for-zero-166899d7","markdown":"/readings/what-should-a-navigator-remember-spatiotemporal-memory-for-zero-166899d7.md","json":"/readings/what-should-a-navigator-remember-spatiotemporal-memory-for-zero-166899d7.json"},"attribution":"Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.","maintainedGuideLinks":[],"readerContext":{"version":"reader-context-v1","attribution":"Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.","terms":[{"id":"term-zero-shot","name":"Zero-shot","explanation":"Doing a task without worked examples (zero-shot) or with only a few examples given in the prompt (few-shot).","referenceUrl":"https://developers.google.com/machine-learning/glossary#zero-shot-prompting"}],"entities":[{"id":"arxiv","name":"arXiv","kind":"organization","context":"arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.","referenceUrls":["https://arxiv.org/","https://info.arxiv.org/help/api/index.html"],"path":"/entities/arxiv","json":"/entities/arxiv.json","markdown":"/entities/arxiv.md"}]},"researchQuestions":[],"withdrawn":false,"references":{"schemaVersion":1,"target":{"type":"reading","id":"166899d74667ced3b5e38c1d51aafb8466b8e310564d1bf8f77d6055212f08f5","revision":"7f4a3397-c633-436b-8496-c65393c4cd32","revisionKind":"publication-id","title":"What Should a Navigator Remember? Spatiotemporal Memory for Zero-Shot Vision-Language Navigation — arXiv cs.AI","url":"https://escapemeridian.com/readings/what-should-a-navigator-remember-spatiotemporal-memory-for-zero-166899d7"},"source":{"type":"source","id":"https://arxiv.org/abs/2610.38318","revision":"e091eb22c08e01ae1a86700997737cd1420e03b19aa17f53f4cc4f2f5d825862","revisionKind":"sha256-source-text","title":"What Should a Navigator Remember? Spatiotemporal Memory for Zero-Shot Vision-Language Navigation","url":"https://arxiv.org/abs/2610.38318"},"status":"available","items":[],"next":null,"scope":"Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter."},"revisions":[{"id":"166899d74667ced3b5e38c1d51aafb8466b8e310564d1bf8f77d6055212f08f5","title":"What Should a Navigator Remember? Spatiotemporal Memory for Zero-Shot Vision-Language Navigation — arXiv cs.AI","argument":"The authors study what an 8B-scale navigation system should retain when following language instructions without navigation-specific training. Their system organizes information into task, temporal, and spatial memories: instruction progress, observation history anchored to poses, and accumulated geometry. An 8B observer summarizes observations; an 8B reasoner uses the memories to choose waypoint intents or request more views, while a geometric controller executes selected targets. The authors distinguish retaining evidence, making it accessible to the reasoner, and acquiring further observations. On R2R-CE val-unseen, the complete system achieved the highest result among the evaluated Qwen3-based zero-shot baselines. Removing temporal memory reduced SR and sharply increased average path length from 16.9 to 39.1m. Removing the spatial-map and planning component also reduced SR. In real-world corridor and office experiments with a Unitree Go2, the zero-shot method achieved higher success rates than a post-trained baseline. The authors say these results suggest better transfer to the evaluated real-world settings. They also use paired instruction interventions in LAW2 to examine instruction-following behavior beyond endpoint success.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The supplied evidence is the abstract, not the full paper; it does not provide the detailed evaluation setup or results underlying the reported comparisons. The real-world transfer result is limited in the abstract to the evaluated corridor and office environments. The abstract reports higher success rates than the post-trained baseline but gives no numerical rates.","speaker":{"name":"Peng, Yuhang, Wang, Zhikai, Wang, Ziyi, Zhong, Jiaru, Zheng, Xiaoji, Huang, Jingzhi, Wang, Yi, Gong, Jiangtao","roleAtSource":"Article author"},"source":{"title":"What Should a Navigator Remember? Spatiotemporal Memory for Zero-Shot Vision-Language Navigation","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2610.38318","publishedAt":"2026-10-28","dateBasis":"Date basis not provided."},"evidence":[{"text":"The system combines task, temporal, and spatial memories for later navigation decisions.","excerpt":"Our agent combines instruction progress, hierarchically compressed observation history anchored to poses, and accumulated geometry in complementary task, temporal, and spatial memories.","url":"https://arxiv.org/abs/2610.38318","relationship":"Source-stated"},{"text":"The system uses an observer to summarize observations and a reasoner to choose targets or request more views, with a controller executing targets.","excerpt":"An 8B observer summarizes observations, and an 8B reasoner uses the memories to select waypoint intents or request additional views, a geometric controller executes the selected targets.","url":"https://arxiv.org/abs/2610.38318","relationship":"Source-stated"},{"text":"On R2R-CE val-unseen, the complete system ranked highest among the evaluated Qwen3-based zero-shot baselines.","excerpt":"On R2R-CE val-unseen, the complete system achieves the highest among the evaluated Qwen3-based zero-shot baselines.","url":"https://arxiv.org/abs/2610.38318","relationship":"Source-stated"},{"text":"Removing temporal memory reduced SR and increased average path length; removing the spatial-map and planning component also reduced SR.","excerpt":"Removing temporal memory reduces SR and sharply increases average path length from 16.9 to 39.1m, removing the spatial-map and planning component reduces SR as well.","url":"https://arxiv.org/abs/2610.38318","relationship":"Source-stated"},{"text":"In evaluated corridor and office experiments, the zero-shot method had higher success rates than the post-trained baseline.","excerpt":"Real-world experiments on a Unitree Go2 show that our zero-shot method achieves higher success rates than the post-trained baseline in both corridor and office environments","url":"https://arxiv.org/abs/2610.38318","relationship":"Source-stated"},{"text":"The authors say their real-world results suggest better transfer to the settings they evaluated.","excerpt":"suggesting better transfer to the evaluated real-world settings.","url":"https://arxiv.org/abs/2610.38318","relationship":"Source-stated"},{"text":"The authors use paired instruction interventions to examine behavioral responses beyond endpoint success.","excerpt":"As a complementary evaluation, LAW2 uses paired instruction interventions to examine behavioral responses, instruction following ability beyond endpoint success.","url":"https://arxiv.org/abs/2610.38318","relationship":"Source-stated"},{"text":"The authors conclude that structured memory preserving task progress, observation history, and geometry supports zero-shot navigation with smaller open models.","excerpt":"These results support preserving task progress, observation history, and geometry in structured memory for zero-shot navigation with smaller open models.","url":"https://arxiv.org/abs/2610.38318","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2610.38318","checkedAt":"2026-11-11T10:42:10.072Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"What evidence from the evaluated corridor and office experiments underlies the reported higher success rates and transfer claim?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"What Should a Navigator Remember? Spatiotemporal Memory for Zero-Shot Vision-Language Navigation — arXiv cs.AI","summary":"The authors study what an 8B-scale navigation system should retain when following language instructions without navigation-specific training. Their system organizes information into task, temporal, and spatial memories: instruction progress, observation history anchored to poses, and accumulated geometry. An 8B observer summarizes observations; an 8B reasoner uses the memories to choose waypoint intents or request more views, while a geometric controller executes selected targets. The authors distinguish retaining evidence, making it accessible to the reasoner, and acquiring further observations. On R2R-CE val-unseen, the complete system achieved the highest result among the evaluated Qwen3-based zero-shot baselines. Removing temporal memory reduced SR and sharply increased average path length from 16.9 to 39.1m. Removing the spatial-map and planning component also reduced SR. In real-world corridor and office experiments with a Unitree Go2, the zero-shot method achieved higher success rates than a post-trained baseline. The authors say these results suggest better transfer to the evaluated real-world settings. They also use paired instruction interventions in LAW2 to examine instruction-following behavior beyond endpoint success.","question":"What evidence from the evaluated corridor and office experiments underlies the reported higher success rates and transfer claim?","claims":[{"text":"The system combines task, temporal, and spatial memories for later navigation decisions.","excerpt":"Our agent combines instruction progress, hierarchically compressed observation history anchored to poses, and accumulated geometry in complementary task, temporal, and spatial memories.","support":"source-stated"},{"text":"The system uses an observer to summarize observations and a reasoner to choose targets or request more views, with a controller executing targets.","excerpt":"An 8B observer summarizes observations, and an 8B reasoner uses the memories to select waypoint intents or request additional views, a geometric controller executes the selected targets.","support":"source-stated"},{"text":"On R2R-CE val-unseen, the complete system ranked highest among the evaluated Qwen3-based zero-shot baselines.","excerpt":"On R2R-CE val-unseen, the complete system achieves the highest among the evaluated Qwen3-based zero-shot baselines.","support":"source-stated"},{"text":"Removing temporal memory reduced SR and increased average path length; removing the spatial-map and planning component also reduced SR.","excerpt":"Removing temporal memory reduces SR and sharply increases average path length from 16.9 to 39.1m, removing the spatial-map and planning component reduces SR as well.","support":"source-stated"},{"text":"In evaluated corridor and office experiments, the zero-shot method had higher success rates than the post-trained baseline.","excerpt":"Real-world experiments on a Unitree Go2 show that our zero-shot method achieves higher success rates than the post-trained baseline in both corridor and office environments","support":"source-stated"},{"text":"The authors say their real-world results suggest better transfer to the settings they evaluated.","excerpt":"suggesting better transfer to the evaluated real-world settings.","support":"source-stated"},{"text":"The authors use paired instruction interventions to examine behavioral responses beyond endpoint success.","excerpt":"As a complementary evaluation, LAW2 uses paired instruction interventions to examine behavioral responses, instruction following ability beyond endpoint success.","support":"source-stated"},{"text":"The authors conclude that structured memory preserving task progress, observation history, and geometry supports zero-shot navigation with smaller open models.","excerpt":"These results support preserving task progress, observation history, and geometry in structured memory for zero-shot navigation with smaller open models.","support":"source-stated"}],"limitations":["The supplied evidence is the abstract, not the full paper; it does not provide the detailed evaluation setup or results underlying the reported comparisons.","The real-world transfer result is limited in the abstract to the evaluated corridor and office environments.","The abstract reports higher success rates than the post-trained baseline but gives no numerical rates."],"mediaId":"a391043ca83168f1c2a6d22029b891db2181efa62fdd497331ca3739fca2aba8","correctionReason":null,"id":"166899d74667ced3b5e38c1d51aafb8466b8e310564d1bf8f77d6055212f08f5","publicationId":"7f4a3397-c633-436b-8496-c65393c4cd32","version":1,"sourceUrl":"https://arxiv.org/abs/2610.38318","sourceTitle":"What Should a Navigator Remember? Spatiotemporal Memory for Zero-Shot Vision-Language Navigation","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-10-28","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-10-30T06:40:04.841Z","createdAt":"2026-11-11T10:11:06.372Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"a391043ca83168f1c2a6d22029b891db2181efa62fdd497331ca3739fca2aba8","kind":"article","url":"https://arxiv.org/abs/2610.38318","sourceUrl":"https://arxiv.org/abs/2610.38318","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-10-31T04:32:22.403Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2610.38318","checkedAt":"2026-11-11T10:42:10.072Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"a391043ca83168f1c2a6d22029b891db2181efa62fdd497331ca3739fca2aba8","confidence":0.6900000000000001,"provider":"typesafe","model":"jev-1.13.0","inputHash":"639a6eff9c3c577ac1df3d0dceded899440e25ac6402b912eaceed23167693e7","decidedAt":"2026-10-31T04:32:23.514Z","redecided":{"from":"stream-admission-v5","under":"stream-admission-v5+confidence-0.5"}},"provenance":{"textHash":"e091eb22c08e01ae1a86700997737cd1420e03b19aa17f53f4cc4f2f5d825862","observationId":"a608345b7327b4e76ee7c2cb0636876ebf842ad1e957fc6b8fb913b8db19845a","retrievedAt":"2026-10-31T04:32:22.231Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"0003fc36-87ee-42d7-8611-8023837d17f7"},"reviewer":{"model":"gpt-6-luna","runId":"d7a7760a-8e6d-42fb-8970-afc84399b92c"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-11T10:11:06.372Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Peng, Yuhang","url":null,"kind":"unknown"},{"name":"Wang, Zhikai","url":null,"kind":"unknown"},{"name":"Wang, Ziyi","url":null,"kind":"unknown"},{"name":"Zhong, Jiaru","url":null,"kind":"unknown"},{"name":"Zheng, Xiaoji","url":null,"kind":"unknown"},{"name":"Huang, Jingzhi","url":null,"kind":"unknown"},{"name":"Wang, Yi","url":null,"kind":"unknown"},{"name":"Gong, Jiangtao","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Peng, Yuhang","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Peng, Yuhang","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.38318","locator":"head/meta[name=author|citation_author]"},"assertionId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-10-31T04:32:22.231Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a","sourceUrl":"https://arxiv.org/abs/2610.38318","contentSha256":"2b26b73da229541cc3dc77ca355d9e7e8f77e838ec9c48be769c1209d7b2820c","textHash":"e091eb22c08e01ae1a86700997737cd1420e03b19aa17f53f4cc4f2f5d825862","observedAt":"2026-10-31T04:32:22.231Z"}},{"actor":{"name":"Wang, Zhikai","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Wang, Zhikai","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.38318","locator":"head/meta[name=author|citation_author]"},"assertionId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-10-31T04:32:22.231Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a","sourceUrl":"https://arxiv.org/abs/2610.38318","contentSha256":"2b26b73da229541cc3dc77ca355d9e7e8f77e838ec9c48be769c1209d7b2820c","textHash":"e091eb22c08e01ae1a86700997737cd1420e03b19aa17f53f4cc4f2f5d825862","observedAt":"2026-10-31T04:32:22.231Z"}},{"actor":{"name":"Wang, Ziyi","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Wang, Ziyi","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.38318","locator":"head/meta[name=author|citation_author]"},"assertionId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-10-31T04:32:22.231Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a","sourceUrl":"https://arxiv.org/abs/2610.38318","contentSha256":"2b26b73da229541cc3dc77ca355d9e7e8f77e838ec9c48be769c1209d7b2820c","textHash":"e091eb22c08e01ae1a86700997737cd1420e03b19aa17f53f4cc4f2f5d825862","observedAt":"2026-10-31T04:32:22.231Z"}},{"actor":{"name":"Zhong, Jiaru","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Zhong, Jiaru","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.38318","locator":"head/meta[name=author|citation_author]"},"assertionId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-10-31T04:32:22.231Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a","sourceUrl":"https://arxiv.org/abs/2610.38318","contentSha256":"2b26b73da229541cc3dc77ca355d9e7e8f77e838ec9c48be769c1209d7b2820c","textHash":"e091eb22c08e01ae1a86700997737cd1420e03b19aa17f53f4cc4f2f5d825862","observedAt":"2026-10-31T04:32:22.231Z"}},{"actor":{"name":"Zheng, Xiaoji","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Zheng, Xiaoji","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.38318","locator":"head/meta[name=author|citation_author]"},"assertionId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-10-31T04:32:22.231Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a","sourceUrl":"https://arxiv.org/abs/2610.38318","contentSha256":"2b26b73da229541cc3dc77ca355d9e7e8f77e838ec9c48be769c1209d7b2820c","textHash":"e091eb22c08e01ae1a86700997737cd1420e03b19aa17f53f4cc4f2f5d825862","observedAt":"2026-10-31T04:32:22.231Z"}},{"actor":{"name":"Huang, Jingzhi","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Huang, Jingzhi","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.38318","locator":"head/meta[name=author|citation_author]"},"assertionId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a:participation:5","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-10-31T04:32:22.231Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a","sourceUrl":"https://arxiv.org/abs/2610.38318","contentSha256":"2b26b73da229541cc3dc77ca355d9e7e8f77e838ec9c48be769c1209d7b2820c","textHash":"e091eb22c08e01ae1a86700997737cd1420e03b19aa17f53f4cc4f2f5d825862","observedAt":"2026-10-31T04:32:22.231Z"}},{"actor":{"name":"Wang, Yi","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Wang, Yi","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.38318","locator":"head/meta[name=author|citation_author]"},"assertionId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a:participation:6","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-10-31T04:32:22.231Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a","sourceUrl":"https://arxiv.org/abs/2610.38318","contentSha256":"2b26b73da229541cc3dc77ca355d9e7e8f77e838ec9c48be769c1209d7b2820c","textHash":"e091eb22c08e01ae1a86700997737cd1420e03b19aa17f53f4cc4f2f5d825862","observedAt":"2026-10-31T04:32:22.231Z"}},{"actor":{"name":"Gong, Jiangtao","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Gong, Jiangtao","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.38318","locator":"head/meta[name=author|citation_author]"},"assertionId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a:participation:7","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-10-31T04:32:22.231Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a","sourceUrl":"https://arxiv.org/abs/2610.38318","contentSha256":"2b26b73da229541cc3dc77ca355d9e7e8f77e838ec9c48be769c1209d7b2820c","textHash":"e091eb22c08e01ae1a86700997737cd1420e03b19aa17f53f4cc4f2f5d825862","observedAt":"2026-10-31T04:32:22.231Z"}}],"participationCoverage":{"total":8,"returned":8,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2610.38318","snapshotId":"ca48e76a6bdff508ae6ddd034282809685671459a43969d90aa1345dc2837f7a","contentSha256":"2b26b73da229541cc3dc77ca355d9e7e8f77e838ec9c48be769c1209d7b2820c","textHash":"e091eb22c08e01ae1a86700997737cd1420e03b19aa17f53f4cc4f2f5d825862","observedAt":"2026-10-31T04:32:22.231Z","basis":"publisher-page-metadata","sourceLinks":47,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/166899d74667ced3b5e38c1d51aafb8466b8e310564d1bf8f77d6055212f08f5/source"}}}],"audits":[{"publicationId":"7f4a3397-c633-436b-8496-c65393c4cd32","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The work is correctly attributed to the paper's authors on arXiv. Results and the transfer interpretation remain framed as the authors' claims. The title suffix says arXiv cs.AI while the primary listing is cs.RO, but this matches the supplied publisher metadata and does not misattribute the work.","excerpt":"Subjects:\nRobotics (cs.RO); Artificial Intelligence (cs.AI)"},{"dimension":"summary","status":"uncertain","explanation":"The summary is faithful, but several central terms are unexplained for a non-specialist: SR (success rate) carries the main ablation result, and R2R-CE val-unseen, Qwen3, 8B, waypoint intents, post-trained baseline, Unitree Go2 and LAW2 lack plain context. It also omits the authors' concluding thesis: structured memory helps zero-shot navigation with smaller open models. Smallest repair: gloss SR and the benchmark, note that Go2 is a robot, and add the thesis sentence.","excerpt":"These results support preserving task progress, observation history, and geometry in structured memory for zero-shot navigation with smaller open models."},{"dimension":"date","status":"pass","explanation":"The originalPublished date of 2026-10-28 matches the v1 submission date. The feed date is later and is correctly not used.","excerpt":"[Submitted on 28 Oct 2026]"},{"dimension":"support","status":"pass","explanation":"Every claim matches the abstract verbatim, including the 16.9 to 39.1m path length. The transfer claim is kept to the evaluated settings and labeled as the authors' suggestion. The limitations correctly note that the abstract gives no numerical real-world rates. The question follows from that gap and does not lead.","excerpt":"suggesting better transfer to the evaluated real-world settings"},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct arXiv abstract page for this paper, which is the most direct way to reach the source.","excerpt":null}],"createdAt":"2026-11-11T10:11:06.372Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}