{"paper":{"title":"Decoupling Exploration and Policy Optimization: Uncertainty Guided Tree Search for Hard Exploration","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Uncertainty-guided tree search decouples exploration from policy optimization to reach SOTA on hard RL benchmarks.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"James Cohan, Zakaria Mhammedi","submitted_at":"2026-03-23T17:56:52Z","abstract_excerpt":"The process of discovery requires active exploration -- the act of collecting new and informative data. However, efficient autonomous exploration remains a major unsolved problem. The dominant paradigm addresses this challenge by using Reinforcement Learning (RL) to train agents with intrinsic motivation, maximizing a composite objective of extrinsic and intrinsic rewards. We suggest that this approach incurs unnecessary overhead: while policy optimization is necessary for precise task execution, employing such machinery solely to expand state coverage may be inefficient. In this paper, we pro"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"By removing the overhead of policy optimization, our approach explores an order of magnitude more efficiently than standard intrinsic motivation baselines on hard exploration benchmarks. ... achieving state-of-the-art performance by a wide margin on Montezuma's Revenge, Pitfall!, and Venture without relying on domain-specific knowledge. ... solving the MuJoCo Adroit dexterous manipulation and AntMaze tasks in a sparse-reward setting, directly from image observations and without expert demonstrations or offline datasets.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That an uncertainty measure paired with Go-With-The-Winner-style tree search will systematically expand state coverage in hard exploration domains without the policy optimization step, and that the resulting trajectories can be reliably distilled into high-performing policies using existing supervised backward learning algorithms.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Uncertainty-guided tree search decouples exploration from RL policy optimization, achieving order-of-magnitude better efficiency and SOTA performance on sparse-reward tasks like Montezuma's Revenge, Pitfall, and Venture via trajectory distillation.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Uncertainty-guided tree search decouples exploration from policy optimization to reach SOTA on hard RL benchmarks.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"ffd20b98b42349f21d1bdbe122ffb09ffa9b2ec1f9508bc22410a3c2cacd0ada"},"source":{"id":"2603.22273","kind":"arxiv","version":4},"verdict":{"id":"30a17c29-74a6-4588-abc3-4c13a76975d2","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T00:17:00.319794Z","strongest_claim":"By removing the overhead of policy optimization, our approach explores an order of magnitude more efficiently than standard intrinsic motivation baselines on hard exploration benchmarks. ... achieving state-of-the-art performance by a wide margin on Montezuma's Revenge, Pitfall!, and Venture without relying on domain-specific knowledge. ... solving the MuJoCo Adroit dexterous manipulation and AntMaze tasks in a sparse-reward setting, directly from image observations and without expert demonstrations or offline datasets.","one_line_summary":"Uncertainty-guided tree search decouples exploration from RL policy optimization, achieving order-of-magnitude better efficiency and SOTA performance on sparse-reward tasks like Montezuma's Revenge, Pitfall, and Venture via trajectory distillation.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That an uncertainty measure paired with Go-With-The-Winner-style tree search will systematically expand state coverage in hard exploration domains without the policy optimization step, and that the resulting trajectories can be reliably distilled into high-performing policies using existing supervised backward learning algorithms.","pith_extraction_headline":"Uncertainty-guided tree search decouples exploration from policy optimization to reach SOTA on hard RL benchmarks."},"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"}