{"paper":{"title":"Proximal Action Replacement for Behavior Cloning Actor-Critic in Offline Reinforcement Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Proximal action replacement overcomes the imitation ceiling in BC-regularized actor-critic by substituting suboptimal dataset actions with value-guided improvements.","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Jianshu Zhang, Jinzong Dong, Nanyang Ye, Qinying Gu, Wei Huang, Xinzhe Yuan, Zhaohui Jiang, Zhuo Chen","submitted_at":"2026-02-07T08:44:27Z","abstract_excerpt":"Offline reinforcement learning (RL), which optimizes policies using a previously collected static dataset, is an important branch of RL. A popular and promising approach is to regularize actor-critic methods with behavior cloning (BC), which quickly yields realistic policies and mitigates bias from out-of-distribution actions, but it can impose an often-overlooked performance ceiling: when dataset actions are suboptimal, indiscriminate imitation structurally prevents the actor from fully exploiting better actions suggested by the value function, especially in later training when imitation is a"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"PAR consistently improves performance across offline RL benchmarks and approaches state-of-the-art results simply by being combined with the basic TD3+BC.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That actions generated by the stable target policy, guided by local ascent of the action-value function and bounded by value uncertainty, can be substituted without destabilizing training or introducing new bias when dataset actions are suboptimal.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Proximal action replacement breaks the imitation ceiling in BC-regularized offline RL actor-critic by substituting suboptimal dataset actions with value-guided improvements from a stable target policy.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Proximal action replacement overcomes the imitation ceiling in BC-regularized actor-critic by substituting suboptimal dataset actions with value-guided improvements.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"1bc0a1fabe3a824f2cbee31b02a28a8816be6ac66e3e322e509f4b137528e168"},"source":{"id":"2602.07441","kind":"arxiv","version":2},"verdict":{"id":"6c7949de-5314-4c6a-a646-019cf8846fd1","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T06:35:10.381452Z","strongest_claim":"PAR consistently improves performance across offline RL benchmarks and approaches state-of-the-art results simply by being combined with the basic TD3+BC.","one_line_summary":"Proximal action replacement breaks the imitation ceiling in BC-regularized offline RL actor-critic by substituting suboptimal dataset actions with value-guided improvements from a stable target policy.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That actions generated by the stable target policy, guided by local ascent of the action-value function and bounded by value uncertainty, can be substituted without destabilizing training or introducing new bias when dataset actions are suboptimal.","pith_extraction_headline":"Proximal action replacement overcomes the imitation ceiling in BC-regularized actor-critic by substituting suboptimal dataset actions with value-guided improvements."},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"b3849650427f66a860cf10c7b6fc96a1e819ae0ca1a3067c74188460cb719665"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}