{"paper":{"title":"SADP: Subgoal-Aware Diffusion Policy for Explainable Robots Learned from Foundation Model Generated Demonstrations","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Conditioning diffusion policies on foundation-model-generated subgoals improves both task success and explainability for robot manipulation.","cross_cats":[],"primary_cat":"cs.RO","authors_text":"Site Hu, Takato Horii","submitted_at":"2026-05-16T08:18:47Z","abstract_excerpt":"Explainable robots require not only successful task execution but also the ability to expose internal decision-making process in a user-friendly manner. However, most imitation learning methods are trained solely on task-level demonstrations, without explicitly modeling subgoal structure or execution progress. This limitation is further exacerbated by the scarcity of subgoal-level supervision in standard robot learning datasets, which restricts the development of robots that can convey the subtasks they are executing during long-horizon manipulation. To address this issue, this paper proposes "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Experiments in RLBench simulations and real-world evaluations on a UR5e robot demonstrate that SADP achieves higher task success rates than strong task-conditioned diffusion baselines, while providing subgoal-level execution signals for monitoring progress and diagnosing failures.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"Foundation models can autonomously generate accurate, unbiased subgoal annotations from raw task demonstrations without introducing systematic errors that would degrade downstream policy learning or explainability.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"SADP trains diffusion policies on foundation-model-generated subgoal-annotated demonstrations and adds a completion predictor to give robots built-in, subgoal-level explainability alongside improved task performance.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Conditioning diffusion policies on foundation-model-generated subgoals improves both task success and explainability for robot manipulation.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"b313d615acee8d7b10c2c749619b964292155dea288f1ceaf04c07558beff267"},"source":{"id":"2605.16871","kind":"arxiv","version":1},"verdict":{"id":"e75007f2-eaf7-49a8-a2af-770e528dfba1","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T20:47:41.639676Z","strongest_claim":"Experiments in RLBench simulations and real-world evaluations on a UR5e robot demonstrate that SADP achieves higher task success rates than strong task-conditioned diffusion baselines, while providing subgoal-level execution signals for monitoring progress and diagnosing failures.","one_line_summary":"SADP trains diffusion policies on foundation-model-generated subgoal-annotated demonstrations and adds a completion predictor to give robots built-in, subgoal-level explainability alongside improved task performance.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"Foundation models can autonomously generate accurate, unbiased subgoal annotations from raw task demonstrations without introducing systematic errors that would degrade downstream policy learning or explainability.","pith_extraction_headline":"Conditioning diffusion policies on foundation-model-generated subgoals improves both task success and explainability for robot manipulation."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.16871/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_compliance","ran_at":"2026-05-19T21:01:24.955218Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T21:01:19.219289Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T18:41:56.297984Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T18:33:26.374379Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"f743495df5d1830143eb052ae1597c02323247593dfd1850c5f54583bbfcb571"},"references":{"count":40,"sample":[{"doi":"","year":2017,"title":"Transparent, explainable, and accountable ai for robotics,","work_id":"1bb30ea2-44a7-4188-9ef2-e2d55235a0bb","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"A review of robot learning for manipulation: Challenges, representations, and algorithms,","work_id":"1b28e85d-e519-4a08-9713-e5b873bd6430","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"A survey of demonstration learning,","work_id":"84d45fbd-7f7b-4f0d-b156-b6095d1ce7eb","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Hierarchical reinforce- ment learning: A survey and open research challenges,","work_id":"de9862f0-13cc-4eb8-9c8c-50b62397917d","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Explainable artificial intelligence: A survey of needs, techniques, applications, and future direction,","work_id":"5cca4f0a-bb88-4509-b90b-189cc9120011","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":40,"snapshot_sha256":"c111967b02a1c8068a19bd1a7087804343cefc1660cb7d9effe5cf11328e8dec","internal_anchors":2},"formal_canon":{"evidence_count":1,"snapshot_sha256":"0c7a65a6f1edd49b9a6983069b038d5d5343fafbb105a61c852e9735de814eef"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}