{"paper":{"title":"GraspVLA: a Grasping Foundation Model Pre-trained on Billion-scale Synthetic Action Data","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"A grasping model pretrained entirely on a billion-frame synthetic dataset achieves open-vocabulary generalization to real robots by unifying perception and action in one chain-of-thought sequence.","cross_cats":[],"primary_cat":"cs.RO","authors_text":"Haixin Ma, Heming Cui, He Wang, Jiayi Chen, Mi Yan, Shengliang Deng, Songlin Wei, Taoyu Yang, Wenhao Zhang, Xuheng Zhang, Yuxin Yang, Zhiqi Zhang, Zhizheng Zhang","submitted_at":"2025-05-06T06:59:28Z","abstract_excerpt":"Embodied foundation models are gaining increasing attention for their zero-shot generalization, scalability, and adaptability to new tasks through few-shot post-training. However, existing models rely heavily on real-world data, which is costly and labor-intensive to collect. Synthetic data offers a cost-effective alternative, yet its potential remains largely underexplored. To bridge this gap, we explore the feasibility of training Vision-Language-Action models entirely with large-scale synthetic action data. We curate SynGrasp-1B, a billion-frame robotic grasping dataset generated in simulat"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"GraspVLA integrates autoregressive perception tasks and flow-matching-based action generation into a unified Chain-of-Thought process, enabling joint training on synthetic action data and Internet semantics data to achieve open-vocabulary generalization in grasping.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That photorealistic rendering plus extensive domain randomization in simulation, combined with the CoT architecture, is sufficient to close the sim-to-real gap so that actions learned on synthetic data transfer effectively to physical robots on unseen objects.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"GraspVLA shows that pretraining a grasping model on a billion synthetic action frames enables zero-shot open-vocabulary performance and sim-to-real transfer.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A grasping model pretrained entirely on a billion-frame synthetic dataset achieves open-vocabulary generalization to real robots by unifying perception and action in one chain-of-thought sequence.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"188944180b282cc6da36a10847252afadf58f39f68815391e810cf3bdbe88d8d"},"source":{"id":"2505.03233","kind":"arxiv","version":3},"verdict":{"id":"9dec9f6d-eea2-43b3-aec4-9d4bdd3e3b29","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-17T20:51:38.996102Z","strongest_claim":"GraspVLA integrates autoregressive perception tasks and flow-matching-based action generation into a unified Chain-of-Thought process, enabling joint training on synthetic action data and Internet semantics data to achieve open-vocabulary generalization in grasping.","one_line_summary":"GraspVLA shows that pretraining a grasping model on a billion synthetic action frames enables zero-shot open-vocabulary performance and sim-to-real transfer.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That photorealistic rendering plus extensive domain randomization in simulation, combined with the CoT architecture, is sufficient to close the sim-to-real gap so that actions learned on synthetic data transfer effectively to physical robots on unseen objects.","pith_extraction_headline":"A grasping model pretrained entirely on a billion-frame synthetic dataset achieves open-vocabulary generalization to real robots by unifying perception and action in one chain-of-thought sequence."},"references":{"count":90,"sample":[{"doi":"","year":2023,"title":"LLaMA: Open and Efficient Foundation Language Models","work_id":"c018fc23-6f3f-4035-9d02-28a2173b2b9d","ref_index":1,"cited_arxiv_id":"2302.13971","is_internal_anchor":true},{"doi":"","year":2023,"title":"Segment Anything","work_id":"2bbf46ca-720a-45a1-8e9c-10c33fbeada0","ref_index":2,"cited_arxiv_id":"2304.02643","is_internal_anchor":true},{"doi":"","year":2021,"title":"Learning Transferable Visual Models From Natural Language Supervision","work_id":"6de86bb5-27bd-4d5c-8b89-967ebfc52659","ref_index":3,"cited_arxiv_id":"2103.00020","is_internal_anchor":true},{"doi":"","year":2023,"title":"Chatgpt: Jan 17 version","work_id":"e752cccc-4880-4947-bff8-1b4c3e4bb107","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control","work_id":"ff438a8a-8003-4fae-9131-acd418b3597b","ref_index":5,"cited_arxiv_id":"2307.15818","is_internal_anchor":true}],"resolved_work":90,"snapshot_sha256":"5cb778e1a6f55163ff9301275a3d0b10174854280aee400a6581d29cf1ba97af","internal_anchors":35},"formal_canon":{"evidence_count":1,"snapshot_sha256":"44a23072815f53252a63d673ee7d458ebdc8aa6fc2aa549dcd376c9f1dd6c8e2"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}