{"paper":{"title":"VIP: Towards Universal Visual Reward and Representation via Value-Implicit Pre-Training","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"VIP pre-trains a visual representation on unlabeled human videos that supplies dense rewards for many robot tasks without any fine-tuning.","cross_cats":["cs.AI","cs.CV","cs.LG"],"primary_cat":"cs.RO","authors_text":"Amy Zhang, Dinesh Jayaraman, Osbert Bastani, Shagun Sodhani, Vikash Kumar, Yecheng Jason Ma","submitted_at":"2022-09-30T18:14:07Z","abstract_excerpt":"Reward and representation learning are two long-standing challenges for learning an expanding set of robot manipulation skills from sensory observations. Given the inherent cost and scarcity of in-domain, task-specific robot data, learning from large, diverse, offline human videos has emerged as a promising path towards acquiring a generally useful visual representation for control; however, how these human videos can be used for general-purpose reward learning remains an open question. We introduce $\\textbf{V}$alue-$\\textbf{I}$mplicit $\\textbf{P}$re-training (VIP), a self-supervised pre-train"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Trained on large-scale Ego4D human videos and without any fine-tuning on in-domain, task-specific data, VIP's frozen representation can provide dense visual reward for an extensive set of simulated and real-robot tasks, enabling diverse reward-based visual control methods and significantly outperforming all prior pre-trained representations.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That a value function learned solely from unlabeled human videos (via an action-free dual goal-conditioned objective) will produce rewards that remain effective when transferred to robotic embodiments and dynamics without further adaptation.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"VIP learns a visual embedding from human videos whose distance defines dense, smooth rewards for arbitrary goal-image robot tasks without task-specific fine-tuning.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"VIP pre-trains a visual representation on unlabeled human videos that supplies dense rewards for many robot tasks without any fine-tuning.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"7081886dfdfbd5fd18d4d774b4cc63125f1bd56d249e8c5522e05a7e35d571be"},"source":{"id":"2210.00030","kind":"arxiv","version":2},"verdict":{"id":"69fd1e82-176a-4ef7-a65a-792e795b7c3e","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T04:38:27.027004Z","strongest_claim":"Trained on large-scale Ego4D human videos and without any fine-tuning on in-domain, task-specific data, VIP's frozen representation can provide dense visual reward for an extensive set of simulated and real-robot tasks, enabling diverse reward-based visual control methods and significantly outperforming all prior pre-trained representations.","one_line_summary":"VIP learns a visual embedding from human videos whose distance defines dense, smooth rewards for arbitrary goal-image robot tasks without task-specific fine-tuning.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That a value function learned solely from unlabeled human videos (via an action-free dual goal-conditioned objective) will produce rewards that remain effective when transferred to robotic embodiments and dynamics without further adaptation.","pith_extraction_headline":"VIP pre-trains a visual representation on unlabeled human videos that supplies dense rewards for many robot tasks without any fine-tuning."},"references":{"count":50,"sample":[{"doi":"","year":null,"title":"Human-to-robot imitation in the wild","work_id":"12b82950-4daa-40b8-bc3f-f94edd67d9db","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"arXiv preprint arXiv:2104.07749 , year=","work_id":"9f8cacca-a18f-412d-9656-7a5274a3531c","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"in-the- wild","work_id":"01177b56-50d1-4b53-8d24-b3f2b8e7809d","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2009,"title":"Imagenet: A large-scale hierarchical image database","work_id":"5d051359-ffec-4827-bbc5-34901db20177","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding","work_id":"ed240a10-5b19-406c-baa5-30803f465785","ref_index":5,"cited_arxiv_id":"1810.04805","is_internal_anchor":true}],"resolved_work":50,"snapshot_sha256":"a461cc54d14940840e47210dbc36444b2a775bc158921bd108db6c52eb80dee8","internal_anchors":12},"formal_canon":{"evidence_count":2,"snapshot_sha256":"1b1ef0bd48678cdddb68d27b987358f05254900d3727703f35c67576ef475588"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}