{"paper":{"title":"TraceVLA: Visual Trace Prompting Enhances Spatial-Temporal Awareness for Generalist Robotic Policies","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Visual trace prompting encodes state-action trajectories to improve spatial-temporal awareness in vision-language-action robotic policies.","cross_cats":["cs.AI"],"primary_cat":"cs.RO","authors_text":"Andrey Kolobov, Furong Huang, Hal Daum\\'e III, Jianfeng Gao, Jianwei Yang, Ruijie Zheng, Shuaiyi Huang, Yongyuan Liang","submitted_at":"2024-12-13T18:40:51Z","abstract_excerpt":"Although large vision-language-action (VLA) models pretrained on extensive robot datasets offer promising generalist policies for robotic learning, they still struggle with spatial-temporal dynamics in interactive robotics, making them less effective in handling complex tasks, such as manipulation. In this work, we introduce visual trace prompting, a simple yet effective approach to facilitate VLA models' spatial-temporal awareness for action prediction by encoding state-action trajectories visually. We develop a new TraceVLA model by finetuning OpenVLA on our own collected dataset of 150K rob"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Evaluations of TraceVLA across 137 configurations in SimplerEnv and 4 tasks on a physical WidowX robot demonstrate state-of-the-art performance, outperforming OpenVLA by 10% on SimplerEnv and 3.5x on real-robot tasks and exhibiting robust generalization across diverse embodiments and scenarios.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the 150K collected trajectories with visual traces are sufficiently diverse and representative so that the observed gains are not artifacts of the specific data-collection procedure or embodiment distribution.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Visual trace prompting improves spatial-temporal awareness in VLA models, delivering 10% gains on SimplerEnv and 3.5x on real-robot tasks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Visual trace prompting encodes state-action trajectories to improve spatial-temporal awareness in vision-language-action robotic policies.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"c77c0fbcb7e18a530d8b2b72a1591ac46f1f42b232f2daeb5a7f84e733169359"},"source":{"id":"2412.10345","kind":"arxiv","version":3},"verdict":{"id":"1b13dae1-e6dc-41f4-8401-2489b6b21245","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T18:22:47.948121Z","strongest_claim":"Evaluations of TraceVLA across 137 configurations in SimplerEnv and 4 tasks on a physical WidowX robot demonstrate state-of-the-art performance, outperforming OpenVLA by 10% on SimplerEnv and 3.5x on real-robot tasks and exhibiting robust generalization across diverse embodiments and scenarios.","one_line_summary":"Visual trace prompting improves spatial-temporal awareness in VLA models, delivering 10% gains on SimplerEnv and 3.5x on real-robot tasks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the 150K collected trajectories with visual traces are sufficiently diverse and representative so that the observed gains are not artifacts of the specific data-collection procedure or embodiment distribution.","pith_extraction_headline":"Visual trace prompting encodes state-action trajectories to improve spatial-temporal awareness in vision-language-action robotic policies."},"references":{"count":76,"sample":[{"doi":"","year":null,"title":"8th Annual Conference on Robot Learning , year=","work_id":"ab0d7be5-0e23-4407-8c37-e64bd7900606","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"7th Annual Conference on Robot Learning , year=","work_id":"259382f5-d2cd-4146-84a3-16a0869c4c73","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"8th Annual Conference on Robot Learning , year=","work_id":"a56f212f-d0d7-45a7-b9a4-c56f0b4f87f0","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Scaling Learning Algorithms Towards","work_id":"bb2761cc-98d0-411b-92f6-803773d64460","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"and Osindero, Simon and Teh, Yee Whye , journal =","work_id":"0a5921e3-ac4e-46f1-85ae-866119a87be0","ref_index":7,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":76,"snapshot_sha256":"6588a7990a8ad8d28c798f8e86c4dd816110b76e5983833c22e065e937bc825c","internal_anchors":16},"formal_canon":{"evidence_count":2,"snapshot_sha256":"fe9bb0d287593b52fc679526f4bd9bb96449dcd9a6371a9e28c1cfd8937ea65b"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}