<strong>AURA: Intent-Directed Probing for Implicit-Need Surfacing in Situated LLM Agents</strong></p>\n<p>Situated queries often carry implicit needs beyond their literal wording — asking <em>\"where is Lin Wei?\"</em> may really be asking <em>\"are they free to interrupt?\"</em> Standard ReAct-style agents take such queries at face value and either miss the underlying need or over-probe with unnecessary tool calls.<br>AURA adds an <strong>IntentFrame</strong> inference step that estimates the implicit need behind a query and computes a <strong>gap score</strong> that directs tool selection — probing only when there's a genuine information gap to close.<br>On a 100-query benchmark, AURA improves coverage of implicit needs over ReAct-style baselines while significantly cutting unnecessary tool probes.<br> Code, simulator, and benchmark: <a href=\"https://github.com/innovation64/AURA\" rel=\"nofollow\">https://github.com/innovation64/AURA</a></p>\n","updatedAt":"2026-06-05T15:21:52.699Z","author":{"_id":"6350c89759bfa9a85d434138","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1666238674117-6350c89759bfa9a85d434138.jpeg","fullname":"Yang Lee","name":"innovation64","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":21,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8387690782546997},"editors":["innovation64"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1666238674117-6350c89759bfa9a85d434138.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2606.05557","authors":[{"_id":"6a22e8f0e4c258a029491713","name":"Yang Li","hidden":false},{"_id":"6a22e8f0e4c258a029491714","name":"Jiaxiang Liu","hidden":false},{"_id":"6a22e8f0e4c258a029491715","name":"Jiang Cai","hidden":false},{"_id":"6a22e8f0e4c258a029491716","name":"Mingkun Xu","hidden":false}],"publishedAt":"2026-06-04T00:00:00.000Z","submittedOnDailyAt":"2026-06-05T00:00:00.000Z","title":"AURA: Intent-Directed Probing for Implicit-Need Surfacing in Situated LLM Agents","submittedOnDailyBy":{"_id":"6350c89759bfa9a85d434138","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1666238674117-6350c89759bfa9a85d434138.jpeg","isPro":false,"fullname":"Yang Lee","user":"innovation64","type":"user","name":"innovation64"},"summary":"A situated query like \"where is Lin Wei?\" often encodes more than its literal content: the user may also want to know whether Lin Wei is free, in a good mood, or worth interrupting now. Standard tool-use agents answer the literal question and stop. AURA inserts an inference step between scene perception and tool use that produces an IntentFrame: a structured estimate of the implicit need with a scalar gap score that controls per-query probe budget and tool selection. On a 100-query four-scene implicit-intent benchmark, AURA improves implicit-need coverage over ReAct-style probing (Delta = +0.07, p < 10^-6); three of four scenes are individually significant, the gain reproduces on a second backbone, and a prompt ablation attributes the lift to gap calibration rather than answer memorisation. On factual lookup the controller trades raw accuracy for 82% fewer probes and zero forbidden-tool violations on a privacy-sensitive slice; scope conditions are detailed in Limitations. Code, simulator, and benchmark are released at https://github.com/innovation64/AURA.","upvotes":1,"discussionId":"6a22e8f0e4c258a029491717","githubRepo":"https://github.com/innovation64/AURA","githubRepoAddedBy":"user","ai_summary":"AURA enhances query answering by incorporating an intent inference step that estimates implicit needs and optimizes tool usage through gap scoring, achieving better implicit-need coverage and reduced probe consumption compared to standard approaches.","ai_keywords":["IntentFrame","gap score","tool-use agents","ReAct-style probing","implicit-need coverage","probe budget","controller","gap calibration"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct","githubStars":0,"organization":{"_id":"6497e1cbe35ff69cfd9b4374","name":"gdiist2021","fullname":"Guangdong Institute of intelligent science and Technology","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/636deed9ffbe479c979d7a2f/6yMAaq3rwb_gf0FiH6lha.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"6350c89759bfa9a85d434138","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1666238674117-6350c89759bfa9a85d434138.jpeg","isPro":false,"fullname":"Yang Lee","user":"innovation64","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"6497e1cbe35ff69cfd9b4374","name":"gdiist2021","fullname":"Guangdong Institute of intelligent science and Technology","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/636deed9ffbe479c979d7a2f/6yMAaq3rwb_gf0FiH6lha.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2606/2606.05557.md"}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2606.05557 in a model README.md to link it from this page.

Cite arxiv.org/abs/2606.05557 in a dataset README.md to link it from this page.

Cite arxiv.org/abs/2606.05557 in a Space README.md to link it from this page.

Add this paper to a collection to link it from this page.