{"paper":{"title":"Action Emergence from Streaming Intent","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Streaming Intent lets an end-to-end driving model generate distinct, high-quality trajectories by deriving and steering with reasoned intent classes.","cross_cats":["cs.CV"],"primary_cat":"cs.RO","authors_text":"Benjin Zhu, Hengtong Lu, Jifeng Dai, Pengfei Jing, Victor Shea-Jay Huang, Xie Yan","submitted_at":"2026-05-12T18:09:04Z","abstract_excerpt":"We formalize action emergence as a target capability for end-to-end autonomous driving: the ability to generate physically feasible, semantically appropriate, and safety-compliant actions in arbitrary, long-tail traffic scenes through scene-conditioned reasoning rather than retrieval or interpolation of learned scene-action mappings. We show that previous paradigms cannot deliver action emergence: autoregressive trajectory decoders collapse the inherently multimodal future into a single averaged output, while diffusion and flow-matching generators express multimodality but are not steerable by"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"SI achieves intent-faithful controllability to our knowledge for the first time in a fully end-to-end VLA: for a fixed scene, varying the intent class at inference yields qualitatively distinct yet consistently high-quality plans arising purely from data-driven learning without any pre-built trajectory bank or hand-coded post-hoc selector.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that the autoregressive chain-of-thought decoding causally derives semantically appropriate intent from scene understanding in a manner that enables generalization to arbitrary long-tail traffic scenes.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A new VLA model called SI uses a four-step chain-of-thought to derive driving intent and applies it via classifier-free guidance to a flow-matching trajectory generator, showing competitive Waymo scores and intent-controllable plans.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Streaming Intent lets an end-to-end driving model generate distinct, high-quality trajectories by deriving and steering with reasoned intent classes.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"2ff0b60625f789855444061395dcb653a0cf013df6b64b5463feae599647adea"},"source":{"id":"2605.12622","kind":"arxiv","version":2},"verdict":{"id":"b02276a1-381f-4b6c-8f21-6aeed293c982","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T05:09:16.671310Z","strongest_claim":"SI achieves intent-faithful controllability to our knowledge for the first time in a fully end-to-end VLA: for a fixed scene, varying the intent class at inference yields qualitatively distinct yet consistently high-quality plans arising purely from data-driven learning without any pre-built trajectory bank or hand-coded post-hoc selector.","one_line_summary":"A new VLA model called SI uses a four-step chain-of-thought to derive driving intent and applies it via classifier-free guidance to a flow-matching trajectory generator, showing competitive Waymo scores and intent-controllable plans.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that the autoregressive chain-of-thought decoding causally derives semantically appropriate intent from scene understanding in a manner that enables generalization to arbitrary long-tail traffic scenes.","pith_extraction_headline":"Streaming Intent lets an end-to-end driving model generate distinct, high-quality trajectories by deriving and steering with reasoned intent classes."},"references":{"count":65,"sample":[{"doi":"","year":null,"title":"Advances in Neural Information Processing Systems , year =","work_id":"83c92541-e6ba-485f-a1f7-db898d79f6d2","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Transactions on Machine Learning Research , year =","work_id":"532ecbf1-56d7-47c5-9913-d815bd63b1b9","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Wod-e2e: Waymo open dataset for end-to-end driving in challenging long-tail scenarios","work_id":"4a3f7018-e4da-4130-8356-de2679f76155","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"AutoVLA: A Vision-Language-Action Model for End-to-End Autonomous Driving with Adaptive Reasoning and Reinforcement Fine-Tuning","work_id":"945172fb-0f3b-43be-88b4-ae61042517e9","ref_index":4,"cited_arxiv_id":"2506.13757","is_internal_anchor":true},{"doi":"","year":null,"title":"arXiv preprint arXiv:2506.11234 (2025)","work_id":"b9c5c925-4544-4d63-aca5-0cabbb44906f","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":65,"snapshot_sha256":"dbdbbd86bb982b58e3e2946575c23938653dc6d0ee2e64aab634fd48796f861e","internal_anchors":9},"formal_canon":{"evidence_count":2,"snapshot_sha256":"9a5252224309d98e3737eb2d44b0941adeb55c51181b6b1312a89bada01addf2"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}