{"paper":{"title":"WarmPrior: Straightening Flow-Matching Policies with Temporal Priors","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Replacing Gaussian noise with recent action history as the source prior straightens flow-matching paths and raises success rates in robot control.","cross_cats":["cs.AI","cs.RO"],"primary_cat":"cs.LG","authors_text":"Chanyoung Kim, Kaixin Wang, Kimin Lee, Li Zhao, Sinjae Kang","submitted_at":"2026-05-13T18:00:01Z","abstract_excerpt":"Generative policies based on diffusion and flow matching have become a dominant paradigm for visuomotor robotic control. We show that replacing the standard Gaussian source distribution with WarmPrior, a simple temporally grounded prior constructed from readily available recent action history, consistently improves success rates on robotic manipulation tasks. We trace this gain to markedly straighter probability paths, echoing the effect of optimal-transport couplings in Rectified Flow. Beyond standard behavior cloning, WarmPrior also reshapes the exploration distribution in prior-space reinfo"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"replacing the standard Gaussian source distribution with WarmPrior, a simple temporally grounded prior constructed from readily available recent action history, consistently improves success rates on robotic manipulation tasks","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That a prior constructed from recent action history will reliably produce straighter paths and performance gains across tasks without introducing harmful temporal biases or reducing robustness to distribution shift.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Replacing Gaussian noise with a temporally grounded prior from recent actions straightens flow-matching paths and improves success rates in robotic manipulation and prior-space RL.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Replacing Gaussian noise with recent action history as the source prior straightens flow-matching paths and raises success rates in robot control.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"98e664fb12fcd0c8c4249a4570fbf2feaa0da21b309769f2cbb09a9023128efd"},"source":{"id":"2605.13959","kind":"arxiv","version":1},"verdict":{"id":"95a22da4-f263-4e5b-8dec-0e459c482195","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T06:10:49.270879Z","strongest_claim":"replacing the standard Gaussian source distribution with WarmPrior, a simple temporally grounded prior constructed from readily available recent action history, consistently improves success rates on robotic manipulation tasks","one_line_summary":"Replacing Gaussian noise with a temporally grounded prior from recent actions straightens flow-matching paths and improves success rates in robotic manipulation and prior-space RL.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That a prior constructed from recent action history will reliably produce straighter paths and performance gains across tasks without introducing harmful temporal biases or reducing robustness to distribution shift.","pith_extraction_headline":"Replacing Gaussian noise with recent action history as the source prior straightens flow-matching paths and raises success rates in robot control."},"references":{"count":15,"sample":[{"doi":"","year":null,"title":"GR00T N1: An Open Foundation Model for Generalist Humanoid Robots","work_id":"e2db69c7-ee8a-4cb7-a761-7b8de1dfcf97","ref_index":1,"cited_arxiv_id":"2503.14734","is_internal_anchor":true},{"doi":"","year":null,"title":"Eagle 2.5: Boosting long-context post-training for frontier vision-language models.arXiv preprint arXiv:2504.15271, 2025a","work_id":"1812f253-bbba-4e3d-96eb-32b2c6481627","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Don’t start from scratch: Behavioral refinement via interpolant-based policy diffusion.arXiv preprint arXiv:2402.16075,","work_id":"d9d5c4b0-4542-43b0-90b3-b19a5a3d38da","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Action-to-Action Flow Matching","work_id":"5a3c672d-3bb1-4ea4-8407-6d0e7f2bbcf9","ref_index":4,"cited_arxiv_id":"2602.07322","is_internal_anchor":true},{"doi":"","year":null,"title":"HAMLET: Switch your Vision-Language-Action Model into a History-Aware Policy","work_id":"4a945e96-4e00-4739-9671-ddd02c7f5cfa","ref_index":5,"cited_arxiv_id":"2510.00695","is_internal_anchor":true}],"resolved_work":15,"snapshot_sha256":"306a40a2bc8f1f2bfcf2592439da58716f9db33396994bc2c2cfbaddee61b9b6","internal_anchors":7},"formal_canon":{"evidence_count":2,"snapshot_sha256":"4c4240e15b9ef04763935e4cc3ee13d02670eb608bf48b6f63f2710ab9ed2083"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}