{"paper":{"title":"Think Twice, Act Once: Verifier-Guided Action Selection For Embodied Agents","license":"http://creativecommons.org/licenses/by/4.0/","headline":"A test-time verifier trained on synthesized failures helps MLLM agents pick reliable actions from multiple candidates.","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Anna Rohrbach, Christian Bialas, Georgia Chalvatzaki, Marcus Rohrbach, Nishad Singhi, Snehal Jauhri, Vignesh Prasad","submitted_at":"2026-05-12T18:08:24Z","abstract_excerpt":"Building generalist embodied agents capable of solving complex real-world tasks remains a fundamental challenge in AI. Multimodal Large Language Models (MLLMs) have significantly advanced the reasoning capabilities of such agents through strong vision-language knowledge and chain-of-thought (CoT) reasoning, yet remain brittle when faced with challenging out-of-distribution scenarios. To address this, we propose Verifier-Guided Action Selection (VegAS), a test-time framework designed to improve the robustness of MLLM-based embodied agents through an explicit verification step. At inference time"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Across embodied reasoning benchmarks spanning the Habitat and ALFRED environments, VeGAS consistently improves generalization, achieving up to a 36% relative performance gain over strong CoT baselines on the most challenging multi-object, long-horizon tasks.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That training a verifier on automatically synthesized failure cases from an LLM will produce a model that reliably identifies good actions in out-of-distribution scenarios where the base MLLM fails.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"VeGAS improves MLLM-based embodied agents by sampling action ensembles and using a verifier trained on LLM-synthesized failure cases, yielding up to 36% relative gains on hard multi-object long-horizon tasks in Habitat and ALFRED.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A test-time verifier trained on synthesized failures helps MLLM agents pick reliable actions from multiple candidates.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"9e87c00acfd66d34ca16915d4712344a09642c67cd415fa10b35d95b88e72510"},"source":{"id":"2605.12620","kind":"arxiv","version":1},"verdict":{"id":"1e46b4c7-8745-451b-91f1-81cf25d3ab8e","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T21:00:48.159839Z","strongest_claim":"Across embodied reasoning benchmarks spanning the Habitat and ALFRED environments, VeGAS consistently improves generalization, achieving up to a 36% relative performance gain over strong CoT baselines on the most challenging multi-object, long-horizon tasks.","one_line_summary":"VeGAS improves MLLM-based embodied agents by sampling action ensembles and using a verifier trained on LLM-synthesized failure cases, yielding up to 36% relative gains on hard multi-object long-horizon tasks in Habitat and ALFRED.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That training a verifier on automatically synthesized failure cases from an LLM will produce a model that reliably identifies good actions in out-of-distribution scenarios where the base MLLM fails.","pith_extraction_headline":"A test-time verifier trained on synthesized failures helps MLLM agents pick reliable actions from multiple candidates."},"references":{"count":65,"sample":[{"doi":"","year":2022,"title":"Do As I Can, Not As I Say: Grounding Language in Robotic Affordances","work_id":"037320f1-b0a9-4cbe-a639-bfb25409ce71","ref_index":1,"cited_arxiv_id":"2204.01691","is_internal_anchor":true},{"doi":"","year":2024,"title":"Critique-out-loud reward models","work_id":"adbb5d79-de43-49c4-abd5-ad004e77abea","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Qwen2.5-VL Technical Report","work_id":"69dffacb-bfe8-442d-be86-48624c60426f","ref_index":3,"cited_arxiv_id":"2502.13923","is_internal_anchor":true},{"doi":"","year":2024,"title":"Large Language Monkeys: Scaling Inference Compute with Repeated Sampling","work_id":"b124064d-5a56-42ad-86f5-3cc349b86a3a","ref_index":4,"cited_arxiv_id":"2407.21787","is_internal_anchor":true},{"doi":"","year":2021,"title":"Training Verifiers to Solve Math Word Problems","work_id":"acab1aa8-b4d6-40e0-a3ee-25341701dca2","ref_index":5,"cited_arxiv_id":"2110.14168","is_internal_anchor":true}],"resolved_work":65,"snapshot_sha256":"c641e77b14dc291d581e8353a882323638234b3b5972594db6be58da7cac2ea4","internal_anchors":7},"formal_canon":{"evidence_count":2,"snapshot_sha256":"ccccfa6436f3acdb87894e5442821afb4e80ad60b5f5b3ed0c56ec6090bc8829"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}