{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:XTWI2FWN7FS74PZUT2ISVUDTQL","short_pith_number":"pith:XTWI2FWN","schema_version":"1.0","canonical_sha256":"bcec8d16cdf965fe3f349e912ad07382c04911c4f88c3404e34c45eecb43e1a5","source":{"kind":"arxiv","id":"2507.21053","version":2},"attestation_state":"computed","paper":{"title":"Flow Matching Policy Gradients","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.RO"],"primary_cat":"cs.LG","authors_text":"Angjoo Kanazawa, Brent Yi, Chung Min Kim, David McAllister, Ethan Weber, Haiwen Feng, Hongsuk Choi, Songwei Ge","submitted_at":"2025-07-28T17:59:57Z","abstract_excerpt":"Flow-based generative models, including diffusion models, excel at modeling continuous distributions in high-dimensional spaces. In this work, we introduce Flow Policy Optimization (FPO), a simple on-policy reinforcement learning algorithm that brings flow matching into the policy gradient framework. FPO casts policy optimization as maximizing an advantage-weighted ratio computed from the conditional flow matching loss, in a manner compatible with the popular PPO-clip framework. It sidesteps the need for exact likelihood computation while preserving the generative capabilities of flow-based mo"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2507.21053","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-07-28T17:59:57Z","cross_cats_sorted":["cs.RO"],"title_canon_sha256":"67d04b473202943761e7bb74c029305ce210fb33830b4ce9503e3cf88c596343","abstract_canon_sha256":"2d0d81c828876aaba20426b234f8bf442c732dec6ae86b528c2f19bbb8afd145"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:46:45.919116Z","signature_b64":"U4qWe5+H97hcgKgjIFEMiwyxmpotfdd1D5DPlQMn7DbbDj7tKc53X2rV2XEZwq0mUAbycKYO+OpB3Z+hG7aDAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bcec8d16cdf965fe3f349e912ad07382c04911c4f88c3404e34c45eecb43e1a5","last_reissued_at":"2026-07-05T11:46:45.918634Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:46:45.918634Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Flow Matching Policy Gradients","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.RO"],"primary_cat":"cs.LG","authors_text":"Angjoo Kanazawa, Brent Yi, Chung Min Kim, David McAllister, Ethan Weber, Haiwen Feng, Hongsuk Choi, Songwei Ge","submitted_at":"2025-07-28T17:59:57Z","abstract_excerpt":"Flow-based generative models, including diffusion models, excel at modeling continuous distributions in high-dimensional spaces. In this work, we introduce Flow Policy Optimization (FPO), a simple on-policy reinforcement learning algorithm that brings flow matching into the policy gradient framework. FPO casts policy optimization as maximizing an advantage-weighted ratio computed from the conditional flow matching loss, in a manner compatible with the popular PPO-clip framework. It sidesteps the need for exact likelihood computation while preserving the generative capabilities of flow-based mo"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.21053","kind":"arxiv","version":2},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2507.21053/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"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"},"aliases":[{"alias_kind":"arxiv","alias_value":"2507.21053","created_at":"2026-07-05T11:46:45.918692+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.21053v2","created_at":"2026-07-05T11:46:45.918692+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.21053","created_at":"2026-07-05T11:46:45.918692+00:00"},{"alias_kind":"pith_short_12","alias_value":"XTWI2FWN7FS7","created_at":"2026-07-05T11:46:45.918692+00:00"},{"alias_kind":"pith_short_16","alias_value":"XTWI2FWN7FS74PZU","created_at":"2026-07-05T11:46:45.918692+00:00"},{"alias_kind":"pith_short_8","alias_value":"XTWI2FWN","created_at":"2026-07-05T11:46:45.918692+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":25,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.21086","citing_title":"ReFPO: Reflow Regularization for Flow Matching Policy Gradients","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2606.15516","citing_title":"Transferring Contact, Not Just Motion: Compliant Grasping Across Dexterous Hands","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2606.13795","citing_title":"DiPOD: Diffusion Policy Optimization without Drifting Apart","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2606.11087","citing_title":"Test-Time Gradient Guidance of Flow Policies in Reinforcement Learning","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2606.06967","citing_title":"GenPO++: Generative Policy Optimization with Jacobian-free Likelihood Ratios","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2606.27475","citing_title":"Support-Constrained RL Enables Real-World Policy Improvement without Real-World Experience","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2605.03065","citing_title":"OGPO: Sample Efficient Full-Finetuning of Generative Control Policies","ref_index":28,"is_internal_anchor":false},{"citing_arxiv_id":"2605.27095","citing_title":"Adversarial Dual On-Policy Distillation from Expressive Teacher","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2605.27736","citing_title":"Explicit Critic Guidance for Aligning Diffusion Models","ref_index":51,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29398","citing_title":"GDSD: Reinforcement Learning as Guided Denoiser Self-Distillation for Diffusion Language Models","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2605.30056","citing_title":"Sample-Efficient Diffusion-based Reinforcement Learning with Critic Guidance","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08879","citing_title":"Preserving Foundational Capabilities in Flow-Matching VLAs through Conservative SFT","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17295","citing_title":"DISA: Offline Importance Sampling for Distribution-Matching LLM-RL","ref_index":50,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19294","citing_title":"DEFLECT: Delay-Robust Execution via Flow-matching Likelihood-Estimated Counterfactual Tuning for VLA Policies","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15458","citing_title":"Video Models Can Reason with Verifiable Rewards","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12625","citing_title":"Driving Intents Amplify Planning-Oriented Reinforcement Learning","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12625","citing_title":"Driving Intents Amplify Planning-Oriented Reinforcement Learning","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12379","citing_title":"Discrete Flow Matching for Offline-to-Online Reinforcement Learning","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08879","citing_title":"Preserving Foundational Capabilities in Flow-Matching VLAs through Conservative SFT","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08733","citing_title":"Generative Actor-Critic with Soft Bridge Policies","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10821","citing_title":"UniSteer: Unified Noise Steering for Efficient Human-Guided VLA Adaptation","ref_index":48,"is_internal_anchor":false},{"citing_arxiv_id":"2604.23380","citing_title":"V-GRPO: Online Reinforcement Learning for Denoising Generative Models Is Easier than You Think","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2604.06916","citing_title":"FP4 Explore, BF16 Train: Diffusion Reinforcement Learning via Efficient Rollout Scaling","ref_index":44,"is_internal_anchor":false},{"citing_arxiv_id":"2604.16519","citing_title":"Positive-Only Drifting Policy Optimization","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2605.03065","citing_title":"OGPO: Sample Efficient Full-Finetuning of Generative Control Policies","ref_index":161,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XTWI2FWN7FS74PZUT2ISVUDTQL","json":"https://pith.science/pith/XTWI2FWN7FS74PZUT2ISVUDTQL.json","graph_json":"https://pith.science/api/pith-number/XTWI2FWN7FS74PZUT2ISVUDTQL/graph.json","events_json":"https://pith.science/api/pith-number/XTWI2FWN7FS74PZUT2ISVUDTQL/events.json","paper":"https://pith.science/paper/XTWI2FWN"},"agent_actions":{"view_html":"https://pith.science/pith/XTWI2FWN7FS74PZUT2ISVUDTQL","download_json":"https://pith.science/pith/XTWI2FWN7FS74PZUT2ISVUDTQL.json","view_paper":"https://pith.science/paper/XTWI2FWN","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.21053&json=true","fetch_graph":"https://pith.science/api/pith-number/XTWI2FWN7FS74PZUT2ISVUDTQL/graph.json","fetch_events":"https://pith.science/api/pith-number/XTWI2FWN7FS74PZUT2ISVUDTQL/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XTWI2FWN7FS74PZUT2ISVUDTQL/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XTWI2FWN7FS74PZUT2ISVUDTQL/action/storage_attestation","attest_author":"https://pith.science/pith/XTWI2FWN7FS74PZUT2ISVUDTQL/action/author_attestation","sign_citation":"https://pith.science/pith/XTWI2FWN7FS74PZUT2ISVUDTQL/action/citation_signature","submit_replication":"https://pith.science/pith/XTWI2FWN7FS74PZUT2ISVUDTQL/action/replication_record"}},"created_at":"2026-07-05T11:46:45.918692+00:00","updated_at":"2026-07-05T11:46:45.918692+00:00"}