{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:22SKRVBN6JPMXE7TRXT7KERWRX","short_pith_number":"pith:22SKRVBN","schema_version":"1.0","canonical_sha256":"d6a4a8d42df25ecb93f38de7f512368de2cb8bbc79db99e0eeb77cc4f1fb830c","source":{"kind":"arxiv","id":"2605.14246","version":1},"attestation_state":"computed","paper":{"title":"Action-Conditioned Risk Gating for Safety-Critical Control under Partial Observability","license":"http://creativecommons.org/licenses/by/4.0/","headline":"A learned action-conditioned predictor of near-term safety violations gates value estimates to approximate risk-sensitive control under partial observability.","cross_cats":["cs.AI","cs.SY","eess.SY"],"primary_cat":"cs.LG","authors_text":"Heng Huang, Tao Wang, Xugui Zhou, Yanfu Zhang, Yin-Jen Chen, Yushen Liu, Ziyi Chen","submitted_at":"2026-05-14T01:23:09Z","abstract_excerpt":"Many safety-critical control problems are modeled as risk-sensitive partially observable Markov decision processes, where the controller must make decisions from incomplete observations while balancing task performance against safety risk. Although belief-space planning provides a principled solution, maintaining and planning over beliefs can be computationally costly and sensitive to model specification in practical domains. We propose a lightweight risk-gated reinforcement learning approximation for risk-sensitive control under partial observability. The method constructs a compact finite-hi"},"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":true,"formal_links_present":false},"canonical_record":{"source":{"id":"2605.14246","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-14T01:23:09Z","cross_cats_sorted":["cs.AI","cs.SY","eess.SY"],"title_canon_sha256":"ea214f43eddd2a9618adef28a0e04e8a22aa30150bcb7982081b855fad359598","abstract_canon_sha256":"3f200740c745531f54a7d7846d87a31d780daab9d32338fab196a8ac00834794"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:39:10.609014Z","signature_b64":"eRod+HirrQxtKKtVRn5qd2NfQ1bqsExmXU8WtzJD9R2duvqDvxl4AS1TaSZ7bf0CQaysDb9esnHlmP34VoUYBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d6a4a8d42df25ecb93f38de7f512368de2cb8bbc79db99e0eeb77cc4f1fb830c","last_reissued_at":"2026-05-17T23:39:10.608575Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:39:10.608575Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Action-Conditioned Risk Gating for Safety-Critical Control under Partial Observability","license":"http://creativecommons.org/licenses/by/4.0/","headline":"A learned action-conditioned predictor of near-term safety violations gates value estimates to approximate risk-sensitive control under partial observability.","cross_cats":["cs.AI","cs.SY","eess.SY"],"primary_cat":"cs.LG","authors_text":"Heng Huang, Tao Wang, Xugui Zhou, Yanfu Zhang, Yin-Jen Chen, Yushen Liu, Ziyi Chen","submitted_at":"2026-05-14T01:23:09Z","abstract_excerpt":"Many safety-critical control problems are modeled as risk-sensitive partially observable Markov decision processes, where the controller must make decisions from incomplete observations while balancing task performance against safety risk. Although belief-space planning provides a principled solution, maintaining and planning over beliefs can be computationally costly and sensitive to model specification in practical domains. We propose a lightweight risk-gated reinforcement learning approximation for risk-sensitive control under partial observability. The method constructs a compact finite-hi"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"The method improves overall glycemic tradeoffs and substantially reduces runtime relative to a belief-space planning baseline; on Safety-Gym it achieves a more favorable reward-cost balance than unconstrained RL and several standard safe-RL baselines.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That a compact finite-history proxy state plus a learned action-conditioned predictor of near-term safety violation is sufficient to produce effective risk-sensitive decisions under partial observability.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Action-conditioned near-term risk prediction gates optimistic and conservative value estimates in RL to approximate risk-sensitive POMDP control, yielding better safety-performance tradeoffs with lower runtime than belief planning baselines.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A learned action-conditioned predictor of near-term safety violations gates value estimates to approximate risk-sensitive control under partial observability.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"da3983a19118da38cb1c1d65d48abc5d362be406f7314533e91fb93b11cceed0"},"source":{"id":"2605.14246","kind":"arxiv","version":1},"verdict":{"id":"dd18a9dc-bf75-4175-aa27-94ce3be932f3","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T02:48:00.155013Z","strongest_claim":"The method improves overall glycemic tradeoffs and substantially reduces runtime relative to a belief-space planning baseline; on Safety-Gym it achieves a more favorable reward-cost balance than unconstrained RL and several standard safe-RL baselines.","one_line_summary":"Action-conditioned near-term risk prediction gates optimistic and conservative value estimates in RL to approximate risk-sensitive POMDP control, yielding better safety-performance tradeoffs with lower runtime than belief planning baselines.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That a compact finite-history proxy state plus a learned action-conditioned predictor of near-term safety violation is sufficient to produce effective risk-sensitive decisions under partial observability.","pith_extraction_headline":"A learned action-conditioned predictor of near-term safety violations gates value estimates to approximate risk-sensitive control under partial observability."},"references":{"count":43,"sample":[{"doi":"","year":2017,"title":"Constrained policy optimization","work_id":"ace446d3-1647-4c52-860a-71a3c7d98d0e","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2017,"title":"Constrained policy optimization","work_id":"dbc1e043-d69a-4f56-9bf8-4642cb64c654","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2017,"title":"A distributional perspective on rein- forcement learning","work_id":"f0303492-1604-45c4-aac5-a66a39390990","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2011,"title":"Wayne Bequette, Darrell M","work_id":"f14324de-0b19-4e09-9997-3338f9d9c47c","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Safe reinforcement learning via shielding under par- tial observability","work_id":"a05af2ae-790a-4f85-bd85-701aeed62ba9","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":43,"snapshot_sha256":"129f3291e446d23a0fe0e0eed55fa3fd41459e003687e0398f51f4ebe05a7a67","internal_anchors":3},"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":"2605.14246","created_at":"2026-05-17T23:39:10.608642+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.14246v1","created_at":"2026-05-17T23:39:10.608642+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.14246","created_at":"2026-05-17T23:39:10.608642+00:00"},{"alias_kind":"pith_short_12","alias_value":"22SKRVBN6JPM","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_16","alias_value":"22SKRVBN6JPMXE7T","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_8","alias_value":"22SKRVBN","created_at":"2026-05-18T12:33:37.589309+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/22SKRVBN6JPMXE7TRXT7KERWRX","json":"https://pith.science/pith/22SKRVBN6JPMXE7TRXT7KERWRX.json","graph_json":"https://pith.science/api/pith-number/22SKRVBN6JPMXE7TRXT7KERWRX/graph.json","events_json":"https://pith.science/api/pith-number/22SKRVBN6JPMXE7TRXT7KERWRX/events.json","paper":"https://pith.science/paper/22SKRVBN"},"agent_actions":{"view_html":"https://pith.science/pith/22SKRVBN6JPMXE7TRXT7KERWRX","download_json":"https://pith.science/pith/22SKRVBN6JPMXE7TRXT7KERWRX.json","view_paper":"https://pith.science/paper/22SKRVBN","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.14246&json=true","fetch_graph":"https://pith.science/api/pith-number/22SKRVBN6JPMXE7TRXT7KERWRX/graph.json","fetch_events":"https://pith.science/api/pith-number/22SKRVBN6JPMXE7TRXT7KERWRX/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/22SKRVBN6JPMXE7TRXT7KERWRX/action/timestamp_anchor","attest_storage":"https://pith.science/pith/22SKRVBN6JPMXE7TRXT7KERWRX/action/storage_attestation","attest_author":"https://pith.science/pith/22SKRVBN6JPMXE7TRXT7KERWRX/action/author_attestation","sign_citation":"https://pith.science/pith/22SKRVBN6JPMXE7TRXT7KERWRX/action/citation_signature","submit_replication":"https://pith.science/pith/22SKRVBN6JPMXE7TRXT7KERWRX/action/replication_record"}},"created_at":"2026-05-17T23:39:10.608642+00:00","updated_at":"2026-05-17T23:39:10.608642+00:00"}