ACT-Agent: Affinity-Cross Transformer for Point Cloud Registration via Reinforcement Learning

ACT-Agent: Affinity-Cross Transformer for Point Cloud Registration via Reinforcement Learning | 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 Research Article ACT-Agent: Affinity-Cross Transformer for Point Cloud Registration via Reinforcement Learning 风光 熊, Haixin Gong, Qiao Ma, Yingbo Jia, Ruize Guo, Yu Cao, Ligang He, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6830295/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Point cloud registration aims to align two sets of point cloud data by determining the geometric transformation between them, specifically rotation and translation. As a fundamental task in 3D computer vision, point cloud registration plays a critical role across various domains. Deep learning-based methods have made significant progress in achieving efficient registration but often lack generalization ability. In this paper, we propose the ACT-Agent registration method, which formulates point cloud registration as a reinforcement learning Markov decision process (MDP) that iteratively optimizes registration to improve accuracy and enhance model robustness. Imitation learning is employed to initialize the registration policy, while a reinforcement learning policy network is used to train an end-to-end model for iterative registration. Furthermore, discrete step sizes are utilized to optimize the execution of individual registration tasks. The model architecture incorporates an Affinity Cross Transformer to extract point cloud features, serving as the state representation within the reinforcement learning environment. The policy network predicts rigid transformation actions accordingly. Soft updates are adopted to improve the optimization efficiency of the reinforcement learning model, coupled with a carefully designed reward function to further boost registration performance. Extensive experiments on ModelNet40 (synthetic data) and ScanObjectNN (real-world data) demonstrate that ACT-Agent achieves higher accuracy, efficiency and generalization ability than the state-of-the-art methods of point cloud registration. point cloud registration reinforcement learning Affinity-Cross Transformer actor critic Full Text Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted 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-6830295","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":467094703,"identity":"74ff989c-249e-4dc3-9439-5829c183d47d","order_by":0,"name":"风光 熊","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA0klEQVRIiWNgGAWjYBACAwglx8APYTATrcWYQbKBZC0GB4jVYs7ee/BxYZuB3ebbPYYfGCqsExvYzx7Aq8Wy51yy8YwzBsnb7pwxlmA4k57YwJOXgN9hN3LMpHkq/iSbARkMjG2HExskeAwIaTH/zWNgALQKpOUfcVrMmHkqDOwMJEBaGojRcuZcsjTPGYMEiRtpxRIJx9KN23hyCGg53nvwM2+bgT3/jOSNHz7UWMv2s5/Br4WBgQdMJjaAyAQgZiOgHq7FnrDCUTAKRsEoGLEAAI+2QK46gbcaAAAAAElFTkSuQmCC","orcid":"","institution":"North University of China","correspondingAuthor":true,"prefix":"","firstName":"风光","middleName":"","lastName":"熊","suffix":""},{"id":467094704,"identity":"49ad423a-82b3-4cad-8678-79acdfc6bd74","order_by":1,"name":"Haixin Gong","email":"","orcid":"","institution":"North University of China","correspondingAuthor":false,"prefix":"","firstName":"Haixin","middleName":"","lastName":"Gong","suffix":""},{"id":467094705,"identity":"fe8feb1c-2afe-4c4c-aba4-0fbae1b299dd","order_by":2,"name":"Qiao Ma","email":"","orcid":"","institution":"North University of China","correspondingAuthor":false,"prefix":"","firstName":"Qiao","middleName":"","lastName":"Ma","suffix":""},{"id":467094706,"identity":"9970ff70-df6d-44e2-bb01-52bfa12a1ac2","order_by":3,"name":"Yingbo Jia","email":"","orcid":"","institution":"North University of China","correspondingAuthor":false,"prefix":"","firstName":"Yingbo","middleName":"","lastName":"Jia","suffix":""},{"id":467094707,"identity":"05414b1e-eab6-4f60-8927-5e1083da2683","order_by":4,"name":"Ruize Guo","email":"","orcid":"","institution":"North University of China","correspondingAuthor":false,"prefix":"","firstName":"Ruize","middleName":"","lastName":"Guo","suffix":""},{"id":467094708,"identity":"3eea670f-fc9f-44fb-ac78-1e95f941e487","order_by":5,"name":"Yu Cao","email":"","orcid":"","institution":"North University of China","correspondingAuthor":false,"prefix":"","firstName":"Yu","middleName":"","lastName":"Cao","suffix":""},{"id":467094709,"identity":"5c757b7b-93ff-463c-8c66-9dc57037f011","order_by":6,"name":"Ligang He","email":"","orcid":"","institution":"University of Warwick","correspondingAuthor":false,"prefix":"","firstName":"Ligang","middleName":"","lastName":"He","suffix":""},{"id":467094710,"identity":"2813e438-9892-4b07-ba04-759f0d2f5f84","order_by":7,"name":"Liqun Kuang","email":"","orcid":"","institution":"North University of China","correspondingAuthor":false,"prefix":"","firstName":"Liqun","middleName":"","lastName":"Kuang","suffix":""},{"id":467094711,"identity":"dfe0c946-2c95-4546-a9a9-6b198c02179c","order_by":8,"name":"Xie Xie","email":"","orcid":"","institution":"North University of China","correspondingAuthor":false,"prefix":"","firstName":"Xie","middleName":"","lastName":"Xie","suffix":""}],"badges":[],"createdAt":"2025-06-05 14:53:15","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6830295/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6830295/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":87208365,"identity":"d6d69f33-ee83-4d60-a516-afaa97ec611c","added_by":"auto","created_at":"2025-07-21 14:23:55","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":816560,"visible":true,"origin":"","legend":"","description":"","filename":"ACTAgent.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6830295/v1_covered_d864ee09-25bc-40ea-b174-25ef94b20832.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"ACT-Agent: Affinity-Cross Transformer for Point Cloud Registration via Reinforcement Learning","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":true,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"point cloud registration, reinforcement learning, Affinity-Cross Transformer, actor critic","lastPublishedDoi":"10.21203/rs.3.rs-6830295/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6830295/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003ePoint cloud registration aims to align two sets of point cloud data by determining the geometric transformation between them, specifically rotation and translation. As a fundamental task in 3D computer vision, point cloud registration plays a critical role across various domains. Deep learning-based methods have made significant progress in achieving efficient registration but often lack generalization ability. In this paper, we propose the ACT-Agent registration method, which formulates point cloud registration as a reinforcement learning Markov decision process (MDP) that iteratively optimizes registration to improve accuracy and enhance model robustness. Imitation learning is employed to initialize the registration policy, while a reinforcement learning policy network is used to train an end-to-end model for iterative registration. Furthermore, discrete step sizes are utilized to optimize the execution of individual registration tasks. The model architecture incorporates an Affinity Cross Transformer to extract point cloud features, serving as the state representation within the reinforcement learning environment. The policy network predicts rigid transformation actions accordingly. Soft updates are adopted to improve the optimization efficiency of the reinforcement learning model, coupled with a carefully designed reward function to further boost registration performance. Extensive experiments on ModelNet40 (synthetic data) and ScanObjectNN (real-world data) demonstrate that ACT-Agent achieves higher accuracy, efficiency and generalization ability than the state-of-the-art methods of point cloud registration.\u003c/p\u003e","manuscriptTitle":"ACT-Agent: Affinity-Cross Transformer for Point Cloud Registration via Reinforcement Learning","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-06-06 07:57:29","doi":"10.21203/rs.3.rs-6830295/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"2bb8995f-0f6a-4b44-96ad-4d2aa7d34d75","owner":[],"postedDate":"June 6th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-07-21T14:23:21+00:00","versionOfRecord":[],"versionCreatedAt":"2025-06-06 07:57:29","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6830295","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6830295","identity":"rs-6830295","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}