{"paper":{"title":"Synthesizing POMDP Policies: Sampling Meets Model-checking via Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Sampling as a membership oracle and model-checking as an equivalence oracle synthesize finite-state POMDP controllers with formal guarantees when the induced policy is regular.","cross_cats":["cs.FL","cs.LO"],"primary_cat":"cs.AI","authors_text":"Anirban Majumdar, Debraj Chakraborty, Jean-Fran\\c{c}ois Raskin, Prince Mathew, Sayan Mukherjee","submitted_at":"2026-05-14T06:37:31Z","abstract_excerpt":"Partially Observable Markov Decision Processes (POMDPs) are the standard framework for decision-making under uncertainty. While sampling-based methods scale well, they lack formal correctness guarantees, making them unsuitable for safety-critical applications. Conversely, formal synthesis techniques provide correctness-by-construction but often struggle with scalability, as general POMDP synthesis is undecidable. To bridge this gap, we propose a synthesis framework that integrates sampling, automata learning, and model-checking. Inspired by Angluin's $L^*$ algorithm, our approach utilizes samp"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We establish a relative completeness result for this framework. Experimental results from our prototypical implementation demonstrate that this method successfully solves threshold-safety problems that remain challenging for existing formal synthesis tools.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The framework provides formal guarantees provided the sampling-induced policy is regular; the approach assumes sampling functions as a reliable membership oracle and model-checking as a reliable equivalence oracle without detailing how noise or approximation in sampling is handled.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A new synthesis framework for POMDPs learns finite-state controllers via sampling and model-checking oracles, achieving relative completeness when the policy is regular and solving threshold-safety problems beyond existing formal tools.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Sampling as a membership oracle and model-checking as an equivalence oracle synthesize finite-state POMDP controllers with formal guarantees when the induced policy is regular.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"8946c918371df7ad5d26dfd6400795cf2414e7c9eef0661ec75bf8b579ef9941"},"source":{"id":"2605.14440","kind":"arxiv","version":1},"verdict":{"id":"6dc38e11-3840-4b02-a046-234609994078","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T01:51:26.470039Z","strongest_claim":"We establish a relative completeness result for this framework. Experimental results from our prototypical implementation demonstrate that this method successfully solves threshold-safety problems that remain challenging for existing formal synthesis tools.","one_line_summary":"A new synthesis framework for POMDPs learns finite-state controllers via sampling and model-checking oracles, achieving relative completeness when the policy is regular and solving threshold-safety problems beyond existing formal tools.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The framework provides formal guarantees provided the sampling-induced policy is regular; the approach assumes sampling functions as a reliable membership oracle and model-checking as a reliable equivalence oracle without detailing how noise or approximation in sampling is handled.","pith_extraction_headline":"Sampling as a membership oracle and model-checking as an equivalence oracle synthesize finite-state POMDP controllers with formal guarantees when the induced policy is regular."},"references":{"count":46,"sample":[{"doi":"","year":2022,"title":"In: Principles of Systems Design: Essays Dedicated to Thomas A","work_id":"129b3120-622c-4d1e-b322-962179e8b33f","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"In: International Con- ference on Computer Aided Verification","work_id":"ae6e13fd-6ccc-43d1-9344-e3c57b862651","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"In: International Conference on Computer Aided Verification","work_id":"73389a62-716f-49a1-81b3-19c9f0e3ed97","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1987,"title":"Angluin, D.: Learning regular sets from queries and counterexamples. Inf. Comput. 75(2), 87–106 (1987)","work_id":"9074eabc-a777-4ac1-b878-8784d5dd7c8a","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2008,"title":"MIT press (2008)","work_id":"ad903039-7863-4cda-a816-8cffb167bd50","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":46,"snapshot_sha256":"dced0dc1fb3b9af95fadafea341254ab42d7696484ea69276a585fee495a284e","internal_anchors":1},"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"}