{"paper":{"title":"Towards Long-horizon Embodied Agents with Tool-Aligned Vision-Language-Action Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Splitting long robot tasks between a high-level planner and specialized action tools raises success rates on extended sequences.","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.RO","authors_text":"Changxing Liu, Minhao Xiong, Siheng Chen, Weixin Li, Yichen Xiong, Yuanzhuo Ding, Zhipeng Zhang, Zixing Lei","submitted_at":"2026-05-13T07:40:34Z","abstract_excerpt":"Vision-language-action (VLA) models are effective robot action executors, but they remain limited on long-horizon tasks due to the dual burden of extended closed-loop planning and diverse physical operations. We therefore propose VLAs-as-Tools, a strategy that distributes this burden across a high-level vision language model (VLM) agent for temporal reasoning and a family of specialized VLA tools for diverse local physical operations. The VLM handles scene analysis, global planning, and recovery, while each VLA tool executes a bounded subtask. To tightly couple agent planning with VLA tool exe"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"VLAs-as-Tools improves the success rate of π_{0.5} by 4.8 points on LIBERO-Long and 23.1 points on RoboTwin, and further enhances invocation fidelity by 15.0 points as measured by Non-biased Rate.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the VLA tool-family interface and Tool-Aligned Post-Training produce specialized tools that reliably follow high-level agent invocations with low error rates and without introducing new failure modes in closed-loop execution.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"VLAs-as-Tools pairs a VLM planner with specialized VLA executors via a new interface and Tool-Aligned Post-Training to raise long-horizon robot success rates on LIBERO-Long and RoboTwin benchmarks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Splitting long robot tasks between a high-level planner and specialized action tools raises success rates on extended sequences.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"6675d8bc269400dcecda1ad6c8f0fe28bfa1f265eb0c8b582f8d63947322e6b3"},"source":{"id":"2605.13119","kind":"arxiv","version":1},"verdict":{"id":"34c5c954-7c5d-44e2-a2f9-650c37e62343","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T18:30:08.903263Z","strongest_claim":"VLAs-as-Tools improves the success rate of π_{0.5} by 4.8 points on LIBERO-Long and 23.1 points on RoboTwin, and further enhances invocation fidelity by 15.0 points as measured by Non-biased Rate.","one_line_summary":"VLAs-as-Tools pairs a VLM planner with specialized VLA executors via a new interface and Tool-Aligned Post-Training to raise long-horizon robot success rates on LIBERO-Long and RoboTwin benchmarks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the VLA tool-family interface and Tool-Aligned Post-Training produce specialized tools that reliably follow high-level agent invocations with low error rates and without introducing new failure modes in closed-loop execution.","pith_extraction_headline":"Splitting long robot tasks between a high-level planner and specialized action tools raises success rates on extended sequences."},"references":{"count":34,"sample":[{"doi":"","year":null,"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":null,"title":"RT-H: Action Hierarchies Using Language","work_id":"ecf7cf18-c1a8-4a6b-bc2a-fb165643aa0d","ref_index":2,"cited_arxiv_id":"2403.01823","is_internal_anchor":true},{"doi":"","year":null,"title":"$\\pi_0$: A Vision-Language-Action Flow Model for General Robot Control","work_id":"f790abdc-a796-482f-a40d-f8ee035ecfc2","ref_index":3,"cited_arxiv_id":"2410.24164","is_internal_anchor":true},{"doi":"","year":null,"title":"RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control","work_id":"ff438a8a-8003-4fae-9131-acd418b3597b","ref_index":4,"cited_arxiv_id":"2307.15818","is_internal_anchor":true},{"doi":"","year":null,"title":"Robo2vlm: Visual question answering from large-scale in-the-wild robot manipulation datasets, 2025a","work_id":"9057ea9c-38d5-4d10-bc50-8c79fb2d9cf2","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":34,"snapshot_sha256":"650cd6eefde82b21bc7d338f5b25f6d4217a8beab094011c0fdfb45e992a4fa2","internal_anchors":20},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}