Co-Training Multimodal World Models and Diffusion-Guided Policies for Zero-Shot Contact- Rich Manipulation

Co-Training Multimodal World Models and Diffusion-Guided Policies for Zero-Shot Contact- Rich Manipulation | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Co-Training Multimodal World Models and Diffusion-Guided Policies for Zero-Shot Contact- Rich Manipulation Lei Li¹, Tianrong Zhang² This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7347334/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 18 You are reading this latest preprint version Abstract This study presents MWM‑DD (Multimodal World Model–Diffusion Decision), a unified control architecture that couples: (i) a vision–tactile–proprioceptive encoder aligned in a shared predictive latent; (ii) a Transformer‑VAE world model that captures latent dynamics and rewards; and (iii) a constraint‑ and value‑guided diffusion decision module realized as a receding‑horizon controller. The system is trained end‑to‑end using reconstruction and Kullback–Leibler (KL) annealing, cross‑modal InfoNCE, and a diffusion ELBO objective, plus kernel‑based off‑policy improvements and domain randomization. Across CALVIN (language‑conditioned rearrangements), FurnitureBench (multi‑stage assemblies), and the Functional Manipulation Benchmark (FMB; single‑ and multi‑object assemblies), the approach achieves state‑of‑the‑art zero‑shot performance and transfers to a Franka Panda + DIGIT platform with small sim‑to‑real gaps. Representative outcomes include 76.7% average simulation success and 69.4% on hardware (20 trials per task), outperforming a strong vision‑only diffusion baseline (53.1% / 45.3%) while maintaining a 7.3‑point sim‑to‑real gap. Ablation results attribute substantial contributions to diffusion (+ 22–27 points), tactile sensing (+ 15–18), cross‑modal alignment (+ 10–13), and zero‑shot regularization (+ 5–10). These findings support the thesis that co‑training predictive latent dynamics with guided diffusion facilitates robust, zero‑shot generalization in contact‑rich manipulation without test‑time fine‑tuning (Chi et al., 2024 ; Mees et al., 2022 ; Heo et al., 2025 ; Luo et al., 2024 ). Physical sciences/Engineering Physical sciences/Mathematics and computing Embodied Robotics Multimodal World Models Diffusion Policy Visuo‑Tactile Fusion Zero‑Shot Generalization Sim‑to‑Real Transfer Full Text Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 29 Oct, 2025 Reviews received at journal 17 Oct, 2025 Reviews received at journal 15 Oct, 2025 Reviews received at journal 11 Oct, 2025 Reviews received at journal 01 Oct, 2025 Reviewers agreed at journal 30 Sep, 2025 Reviewers agreed at journal 27 Sep, 2025 Reviewers agreed at journal 26 Sep, 2025 Reviewers agreed at journal 26 Sep, 2025 Reviewers agreed at journal 25 Sep, 2025 Reviewers agreed at journal 25 Sep, 2025 Reviews received at journal 25 Sep, 2025 Reviewers agreed at journal 25 Sep, 2025 Reviewers invited by journal 22 Sep, 2025 Editor invited by journal 20 Aug, 2025 Editor assigned by journal 19 Aug, 2025 Submission checks completed at journal 19 Aug, 2025 First submitted to journal 11 Aug, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7347334","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":523372966,"identity":"d1c72ecb-a1be-4671-b9b7-96fd8839f095","order_by":0,"name":"Lei Li¹","email":"","orcid":"","institution":"¹Hangzhou City University Procurement Center","correspondingAuthor":false,"prefix":"","firstName":"Lei","middleName":"","lastName":"Li¹","suffix":""},{"id":523372967,"identity":"76ecbbb1-d1ac-44a8-adbd-7b8c9554ba6b","order_by":1,"name":"Tianrong Zhang²","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA6ElEQVRIie3QPQrCMBTA8VcCmQKuET96BSVQF/EsTwS76CSIY6W7c0XBK3RyDgh26RlEETo5CKJ000RxTXUTzH9IMrwfCQGw2X4x/trqAPJ5cILPCIL4nnTfk8XEXYTZ4Zrv/NU0yS4M2rVYkmxvIs5y2xJVHA2nQSoWDPoilrTVMBHC0atwxGEIKSEMNt1YMspNhHL/polPITkqci8mjA+88hkRGUihiCwmnA/GFehjM4JUOMtGT8w31DMSN/LX5byNrhslRzhNOrVZEmZGolPv0fdJteivIkXzKifXayn4YNRms9n+sgcF3UOUEySc8gAAAABJRU5ErkJggg==","orcid":"","institution":"¹Hangzhou City University Procurement Center","correspondingAuthor":true,"prefix":"","firstName":"Tianrong","middleName":"","lastName":"Zhang²","suffix":""}],"badges":[],"createdAt":"2025-08-11 14:23:27","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7347334/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7347334/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":92693522,"identity":"d8f5705b-1051-4476-bfff-2dd8335c7473","added_by":"auto","created_at":"2025-10-03 06:14:45","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1095482,"visible":true,"origin":"","legend":"","description":"","filename":"0817authorZeroShotContactRichManipulationManuscript.docx","url":"https://assets-eu.researchsquare.com/files/rs-7347334/v1/522bf381cca6329164762e1e.docx"},{"id":92693748,"identity":"15811d8d-911a-4f99-baf6-21bd952b3551","added_by":"auto","created_at":"2025-10-03 06:22:45","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4425,"visible":true,"origin":"","legend":"","description":"","filename":"25d3583151984e17bd8f7dbbc426e488.json","url":"https://assets-eu.researchsquare.com/files/rs-7347334/v1/9a67d4cf3bce1e1588208683.json"},{"id":92693518,"identity":"8a7f083b-e74e-4461-90e3-db2072c291c7","added_by":"auto","created_at":"2025-10-03 06:14:45","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":118018,"visible":true,"origin":"","legend":"","description":"","filename":"25d3583151984e17bd8f7dbbc426e4881enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7347334/v1/fe5020e7230c983b0530d8e5.xml"},{"id":92694342,"identity":"aa39a838-eab5-4023-8b7a-5e741bd6233e","added_by":"auto","created_at":"2025-10-03 06:30:45","extension":"png","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":394194,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7347334/v1/9ef4cde67150d5dc1c2361e5.png"},{"id":92693516,"identity":"6dd2b232-94a7-4211-96e6-788ce6d70934","added_by":"auto","created_at":"2025-10-03 06:14:45","extension":"png","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":114158,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7347334/v1/dbdf1929d32b7937ac335365.png"},{"id":92693515,"identity":"9c4f9955-14fd-428c-b022-9eca21ccd55a","added_by":"auto","created_at":"2025-10-03 06:14:45","extension":"png","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":156855,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7347334/v1/17d42726aa13ffaeec5d3103.png"},{"id":92693749,"identity":"ea1e07df-3bbf-444f-96a2-3c83b6976309","added_by":"auto","created_at":"2025-10-03 06:22:45","extension":"png","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":372599,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7347334/v1/0a0f403e7f63bc5608a7c36a.png"},{"id":92693521,"identity":"95022a62-e74f-4e15-91bc-8215e0856ef8","added_by":"auto","created_at":"2025-10-03 06:14:45","extension":"png","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":23208,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7347334/v1/b315807ed1c0b53ba384caab.png"},{"id":92693524,"identity":"39c6ba00-4ffd-4f01-b3e5-a57bd043a3c9","added_by":"auto","created_at":"2025-10-03 06:14:46","extension":"png","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":27463,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7347334/v1/d9bf2bf4ad1bd92ac4486439.png"},{"id":92693751,"identity":"405b66cc-1699-42b4-a86e-688762a91b8e","added_by":"auto","created_at":"2025-10-03 06:22:46","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":29289,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7347334/v1/d01aaf78980db57c6633401c.png"},{"id":92693523,"identity":"d42bc681-462c-43fb-9c07-5b16e6c29589","added_by":"auto","created_at":"2025-10-03 06:14:46","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":24288,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7347334/v1/4b37bf9e5a753edeef40e0e0.png"},{"id":92693527,"identity":"540b7b6a-8867-416e-b12c-a00330650f9a","added_by":"auto","created_at":"2025-10-03 06:14:46","extension":"xml","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":118621,"visible":true,"origin":"","legend":"","description":"","filename":"25d3583151984e17bd8f7dbbc426e4881structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7347334/v1/cf73e958809d9f2902346cb8.xml"},{"id":92693526,"identity":"63147e6b-31f3-45bb-9004-d7845c709dec","added_by":"auto","created_at":"2025-10-03 06:14:46","extension":"html","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":135889,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7347334/v1/d30c1516d2101cdb69b9d221.html"},{"id":92694891,"identity":"a947b617-909e-4e9f-8e65-bb5cf51d8de0","added_by":"auto","created_at":"2025-10-03 06:38:59","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":676911,"visible":true,"origin":"","legend":"","description":"","filename":"0817authorZeroShotContactRichManipulationManuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7347334/v1_covered_a99ec587-e536-4d30-9d52-40bc50510fdf.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Co-Training Multimodal World Models and Diffusion-Guided Policies for Zero-Shot Contact- Rich Manipulation","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":true,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Embodied Robotics, Multimodal World Models, Diffusion Policy, Visuo‑Tactile Fusion, Zero‑Shot Generalization, Sim‑to‑Real Transfer","lastPublishedDoi":"10.21203/rs.3.rs-7347334/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7347334/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThis study presents \u003cb\u003eMWM‑DD\u003c/b\u003e (Multimodal World Model\u0026ndash;Diffusion Decision), a unified control architecture that couples: (i) a vision\u0026ndash;tactile\u0026ndash;proprioceptive encoder aligned in a shared predictive latent; (ii) a Transformer‑VAE world model that captures latent dynamics and rewards; and (iii) a constraint‑ and value‑guided \u003cb\u003ediffusion decision\u003c/b\u003e module realized as a receding‑horizon controller. The system is trained end‑to‑end using reconstruction and Kullback\u0026ndash;Leibler (KL) annealing, cross‑modal InfoNCE, and a diffusion ELBO objective, plus kernel‑based off‑policy improvements and domain randomization. Across CALVIN (language‑conditioned rearrangements), FurnitureBench (multi‑stage assemblies), and the Functional Manipulation Benchmark (FMB; single‑ and multi‑object assemblies), the approach achieves state‑of‑the‑art \u003cb\u003ezero‑shot\u003c/b\u003e performance and transfers to a Franka Panda\u0026thinsp;+\u0026thinsp;DIGIT platform with small sim‑to‑real gaps. Representative outcomes include 76.7% average simulation success and 69.4% on hardware (20 trials per task), outperforming a strong vision‑only diffusion baseline (53.1% / 45.3%) while maintaining a 7.3‑point sim‑to‑real gap. Ablation results attribute substantial contributions to diffusion (+\u0026thinsp;22\u0026ndash;27 points), tactile sensing (+\u0026thinsp;15\u0026ndash;18), cross‑modal alignment (+\u0026thinsp;10\u0026ndash;13), and zero‑shot regularization (+\u0026thinsp;5\u0026ndash;10). These findings support the thesis that \u003cb\u003eco‑training predictive latent dynamics with guided diffusion\u003c/b\u003e facilitates robust, zero‑shot generalization in contact‑rich manipulation without test‑time fine‑tuning (Chi et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Mees et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Heo et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Luo et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e","manuscriptTitle":"Co-Training Multimodal World Models and Diffusion-Guided Policies for Zero-Shot Contact- Rich Manipulation","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-10-03 06:14:41","doi":"10.21203/rs.3.rs-7347334/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-10-29T08:48:42+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-10-17T05:41:53+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-10-15T20:02:13+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-10-11T15:40:38+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-10-01T12:25:40+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"205608053382852543207964463128111101856","date":"2025-09-30T10:06:39+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"35094695517959108648888920068636008822","date":"2025-09-27T04:11:12+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"199828691214753858359843055313454047863","date":"2025-09-27T02:22:20+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"153264999011204264033786591302647149321","date":"2025-09-26T20:51:02+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"312265241015425528370037168984338071984","date":"2025-09-25T11:18:23+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"21065239040190460523907637528949034078","date":"2025-09-25T05:27:33+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-09-25T04:50:15+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"322550641901014534883426703768729191208","date":"2025-09-25T04:39:23+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-09-22T09:46:55+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-08-20T14:18:10+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-08-19T08:45:27+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-08-19T04:12:02+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2025-08-11T14:19:47+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"774ed82e-0508-4eaa-bb5f-91451e096b8e","owner":[],"postedDate":"October 3rd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":55631329,"name":"Physical sciences/Engineering"},{"id":55631330,"name":"Physical sciences/Mathematics and computing"}],"tags":[],"updatedAt":"2026-01-16T11:42:10+00:00","versionOfRecord":[],"versionCreatedAt":"2025-10-03 06:14:41","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7347334","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7347334","identity":"rs-7347334","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}