{ "schema_version": 1, "builder_plan": { "scenario": "An assistant chooses the wrong API or sends arguments your product cannot use.", "change": "Apply the proposed tool-selection or argument-generation change in a sandbox.", "baseline": "Your current agent with the same tool schemas and task budget.", "steps": [ "Locate the paper's implementation and check its license, model support, data needs, and hardware requirements. Stop here if you cannot obtain the required artifacts.", "Save a small set of representative product tasks and expected outcomes. Keep evaluation tasks out of any training or tuning, and choose quality and cost limits before running.", "Run the baseline and the proposed change on identical inputs. Keep other settings fixed and repeat paired runs to expose variation.", "Save per-task outputs, failures, timings, and all usage costs. Review regressions before expanding the test." ], "measure": "Correct completed actions, invalid calls, retries, latency, and cost per completed task.", "decision": "Keep the change only if it meets your preselected quality and latency limits and improves the metric you care about after setup and operating costs. Otherwise retain the baseline. This is an exploration plan, not a validated recommendation." }, "builder_review": null, "arxiv_id": "2610.18356", "title": "Urbex: Agentic Spatial Grounding for City-Scale 3D Scenes", "summary": "Urbex replaces candidate-ranking pipelines that depend on precomputed 3D instance proposals with a VLM agent that actively navigates a multi-view city scene (querying landmarks, zooming, rendering oblique views) before committing a bounding box, trained with reinforcement learning on CityRefer. The cost is an interactive multi-view rendering environment, RL training of the tool-use policy with localization and shaping rewards, and a VLM agent capable of sequential spatial reasoning, none of which is off-the-shelf for a constrained-infrastructure team. Failure modes include the agent committing to a wrong region during search, degraded performance on scenes whose geometry or landmark distribution differs from CityRefer, and environment/rendering engineering overhead that may exceed the cost of a simpler ranking baseline for smaller scenes.", "rationale": "The method targets a specialized city-scale 3D grounding setting whose interactive environment, RL training data, and VLM agent requirements are not practically reproducible on a 24 GB GPU with a small team, though it is worth monitoring as an alternative to proposal-based pipelines.", "technique": "Agentic spatial search for city-scale 3D grounding", "family": "uso_de_ferramenta", "recommendation": "observar", "published": "2026-10-13", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.18356", "independent_implementations": 2, "total_implementations": 2, "stars": 76, "citations": null, "claims": [ { "claim_id": "claim-01", "statement": "Urbex replaces candidate-ranking pipelines that depend on precomputed 3D instance proposals with a VLM agent that actively navigates a multi-view city scene (querying landmarks, zooming, rendering oblique views) before committing a bounding box, trained with reinforcement learning on CityRefer. The cost is an interactive multi-view rendering environment, RL training of the tool-use policy with localization and shaping rewards, and a VLM agent capable of sequential spatial reasoning, none of which is off-the-shelf for a constrained-infrastructure team. Failure modes include the agent committing to a wrong region during search, degraded performance on scenes whose geometry or landmark distribution differs from CityRefer, and environment/rendering engineering overhead that may exceed the cost of a simpler ranking baseline for smaller scenes.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.18356", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Best Hit@1 and IoU on CityRefer, substantially faster than the strongest ranking baseline, with improved zero-shot robustness on STPLS3D-Refer; no numeric factor is stated.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.18356", "source_page": null, "source_excerpt": "" } ], "exposure_map": [ { "dimension": "quality", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "compute", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "latency", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "operations", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "compatibility", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "security", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "data_and_training", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "reproducibility", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" } ], "risks": [], "minimum_test": [], "open_questions": [], "independent_tests": [], "lab_reproductions": [], "report_available": false, "validation_tier": "unknown", "evidence_tier": "unknown", "infrastructure_basis": "unknown", "equations": [], "equations_status": "not_fetched", "core_kind": "", "equations_fetched_at": "", "equations_source_url": "", "equations_source_sha256": "", "exposures_reviewed_at": "", "exposures_source_sha256": "" }