{"paper":{"title":"Dream-MPC: Gradient-Based Model Predictive Control with Latent Imagination","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"Dream-MPC refines a few policy-rollout trajectories via gradient ascent inside a learned world model to raise overall task performance.","cross_cats":["cs.AI","cs.RO"],"primary_cat":"cs.LG","authors_text":"Jonathan Spieler, Sven Behnke","submitted_at":"2026-05-06T07:13:11Z","abstract_excerpt":"State-of-the-art model-based Reinforcement Learning (RL) approaches either use gradient-free, population-based methods for planning, learned policy networks, or a combination of policy networks and planning. Hybrid approaches that combine Model Predictive Control (MPC) with a learned model and a policy prior to leverage the advantages of both paradigms have shown promising results. However, these approaches typically rely on gradient-free optimization methods, which can be computationally expensive for high-dimensional control tasks. While gradient-based methods are a promising alternative, re"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Our results on 24 continuous control tasks show that Dream-MPC can significantly improve the performance of the underlying policy and can outperform gradient-free MPC and state-of-the-art baselines.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The learned world model is sufficiently accurate and differentiable that gradient ascent on rolled-out trajectories reliably improves performance without being misled by model errors or local optima.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Dream-MPC boosts underlying policies on 24 continuous control tasks by optimizing policy-generated trajectories with gradient ascent, uncertainty regularization, and temporal amortization inside a latent world model.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Dream-MPC refines a few policy-rollout trajectories via gradient ascent inside a learned world model to raise overall task performance.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"12fdb68e98465d566cdab351d22705ffba0aa13bc225bbcf49e8ca24164555df"},"source":{"id":"2605.04568","kind":"arxiv","version":2},"verdict":{"id":"3e10bde3-1524-4c77-a33e-565509147905","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-08T16:27:37.476846Z","strongest_claim":"Our results on 24 continuous control tasks show that Dream-MPC can significantly improve the performance of the underlying policy and can outperform gradient-free MPC and state-of-the-art baselines.","one_line_summary":"Dream-MPC boosts underlying policies on 24 continuous control tasks by optimizing policy-generated trajectories with gradient ascent, uncertainty regularization, and temporal amortization inside a latent world model.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The learned world model is sufficiently accurate and differentiable that gradient ascent on rolled-out trajectories reliably improves performance without being misled by model errors or local optima.","pith_extraction_headline":"Dream-MPC refines a few policy-rollout trajectories via gradient ascent inside a learned world model to raise overall task performance."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.04568/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-20T11:39:47.694956Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T22:31:20.097591Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T14:20:08.650861Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"7a5f27b019c08312cefe236fc31addcf375473790e0ef05a1917df7b1cec37de"},"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"}