[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"global-home":3,"{\"version\":\"published\"}journal":529,"site-settings":606,"blog-posts-published":611},{"data":4,"headers":506},{"story":5,"cv":503,"rels":504,"links":505},{"name":6,"created_at":7,"published_at":8,"updated_at":9,"id":10,"uuid":11,"content":12,"slug":495,"full_slug":495,"sort_by_date":218,"position":496,"tag_list":497,"is_startpage":92,"parent_id":498,"meta_data":218,"group_id":499,"first_published_at":500,"release_id":218,"lang":501,"path":218,"alternates":502,"default_full_slug":218,"translated_slugs":218},"Home","2026-02-24T17:14:26.572Z","2026-04-22T17:57:28.496Z","2026-04-22T17:57:28.511Z",148535978303585,"d77fee8c-f275-43b5-9639-1b7c235c5407",{"_uid":13,"body":14,"_editable":491,"component":492,"seo_title":493,"coming_soon":92,"seo_description":494},"0285b3db-5933-4051-967a-420df86db660",[15,56,96,120,234,257,408,461,481],{"_uid":16,"visible":17,"cta_link":18,"cta_soon":17,"cta_text":21,"headline":22,"component":23,"cta_color":24,"description":25,"subheadline":39,"partner_logos":51,"cta_secondary_link":52,"cta_secondary_soon":17,"cta_secondary_text":54,"cta_secondary_color":55},"5bedde42",true,{"linktype":19,"cached_url":20},"url","\u002Fmarketplace","Browse our marketplace","Human-Created Datasets","hero_video","salmon",{"type":26,"content":27},"doc",[28],{"type":29,"content":30},"paragraph",[31,37],{"text":32,"type":33,"marks":34},"High-quality,","text",[35],{"type":36},"bold",{"text":38,"type":33}," affordable, on-demand robotics data collection built for teams that need to move fast, experiment freely, and scale confidently.",{"type":26,"content":40},[41],{"type":29,"content":42},[43,45,49],{"text":44,"type":33},"To Scale Your ",{"text":46,"type":33,"marks":47},"Robotics",[48],{"type":36},{"text":50,"type":33}," Journey",[],{"linktype":19,"cached_url":53},"\u002Fbespoke-data","Request a bespoke solution","",{"_uid":57,"links":58,"visible":17,"cta_link":93,"cta_text":55,"_editable":94,"component":95},"nav-001",[59,65,70,75,81,87],{"_uid":60,"link":61,"label":6,"_editable":63,"component":64},"link-001",{"linktype":19,"cached_url":62},"\u002F","\u003C!--#storyblok#{\"name\": \"nav_link\", \"space\": \"290758475334152\", \"uid\": \"link-001\", \"id\": \"148535978303585\"}-->","nav_link",{"_uid":66,"link":67,"label":68,"_editable":69,"component":64},"link-002",{"linktype":19,"cached_url":20},"Marketplace","\u003C!--#storyblok#{\"name\": \"nav_link\", \"space\": \"290758475334152\", \"uid\": \"link-002\", \"id\": \"148535978303585\"}-->",{"_uid":71,"link":72,"label":73,"_editable":74,"component":64},"link-003",{"linktype":19,"cached_url":53},"Bespoke-Data","\u003C!--#storyblok#{\"name\": \"nav_link\", \"space\": \"290758475334152\", \"uid\": \"link-003\", \"id\": \"148535978303585\"}-->",{"_uid":76,"link":77,"label":79,"_editable":80,"component":64},"link-004",{"linktype":19,"cached_url":78},"\u002Fhardware","Hardware","\u003C!--#storyblok#{\"name\": \"nav_link\", \"space\": \"290758475334152\", \"uid\": \"link-004\", \"id\": \"148535978303585\"}-->",{"_uid":82,"link":83,"label":85,"_editable":86,"component":64},"link-005",{"linktype":19,"cached_url":84},"\u002Fabout","About Us","\u003C!--#storyblok#{\"name\": \"nav_link\", \"space\": \"290758475334152\", \"uid\": \"link-005\", \"id\": \"148535978303585\"}-->",{"_uid":88,"link":89,"label":91,"component":64,"coming_soon":92},"677f2713",{"linktype":19,"cached_url":90},"\u002Fjournal","Journal",false,{"linktype":19,"cached_url":55},"\u003C!--#storyblok#{\"name\": \"navbar\", \"space\": \"290758475334152\", \"uid\": \"nav-001\", \"id\": \"148535978303585\"}-->","navbar",{"_uid":97,"visible":17,"cta_link":98,"cta_text":21,"headline":22,"_editable":99,"component":100,"show_waves":17,"description":101,"subheadline":109,"partner_logos":118,"cta_secondary_link":119,"cta_secondary_text":54,"cta_secondary_color":55},"hero-001",{"linktype":19,"cached_url":20},"\u003C!--#storyblok#{\"name\": \"hero\", \"space\": \"290758475334152\", \"uid\": \"hero-001\", \"id\": \"148535978303585\"}-->","hero",{"type":26,"content":102},[103],{"type":29,"content":104},[105,108],{"text":32,"type":33,"marks":106},[107],{"type":36},{"text":38,"type":33},{"type":26,"content":110},[111],{"type":29,"content":112},[113,114,117],{"text":44,"type":33},{"text":46,"type":33,"marks":115},[116],{"type":36},{"text":50,"type":33},[],{"linktype":19,"cached_url":53},{"_uid":121,"cards":122,"visible":17,"headline":188,"subtitle":198,"_editable":212,"component":213,"description":214},"pain-001",[123,140,156,172],{"_uid":124,"title":125,"_editable":126,"component":127,"description":128},"pc-001","Not enough data","\u003C!--#storyblok#{\"name\": \"pain_card\", \"space\": \"290758475334152\", \"uid\": \"pc-001\", \"id\": \"148535978303585\"}-->","pain_card",{"type":26,"content":129},[130],{"type":29,"content":131},[132,134,138],{"text":133,"type":33},"Your models need ",{"text":135,"type":33,"marks":136},"thousands of diverse trajectories and hours of real interaction.",[137],{"type":36},{"text":139,"type":33}," But this kind of data doesn't really exist on the market yet. So every new task feels like you're building the first dataset from scratch.",{"_uid":141,"title":142,"_editable":143,"component":127,"description":144},"pc-002","Data that doesn't fit your needs","\u003C!--#storyblok#{\"name\": \"pain_card\", \"space\": \"290758475334152\", \"uid\": \"pc-002\", \"id\": \"148535978303585\"}-->",{"type":26,"content":145},[146],{"type":29,"content":147},[148,150,154],{"text":149,"type":33},"You find datasets, but they're not built for your ",{"text":151,"type":33,"marks":152},"embodiment, DOFs, or sensor type.",[153],{"type":36},{"text":155,"type":33}," You spend cycles hacking around mismatched data instead of training the model you really want.",{"_uid":157,"title":158,"_editable":159,"component":127,"description":160},"pc-003","Slow and expensive to collect","\u003C!--#storyblok#{\"name\": \"pain_card\", \"space\": \"290758475334152\", \"uid\": \"pc-003\", \"id\": \"148535978303585\"}-->",{"type":26,"content":161},[162],{"type":29,"content":163},[164,166,170],{"text":165,"type":33},"Building your own infrastructure is ",{"text":167,"type":33,"marks":168},"costly.",[169],{"type":36},{"text":171,"type":33}," Each iteration means weeks of new data collection and delays.",{"_uid":173,"title":174,"_editable":175,"component":127,"description":176},"pc-004","Your team wants to train, not manage data ops","\u003C!--#storyblok#{\"name\": \"pain_card\", \"space\": \"290758475334152\", \"uid\": \"pc-004\", \"id\": \"148535978303585\"}-->",{"type":26,"content":177},[178],{"type":29,"content":179},[180,182,186],{"text":181,"type":33},"Your team wants to focus on training and testing. Instead, they spend a big part of their week ",{"text":183,"type":33,"marks":184},"planning, collecting, and cleaning data",[185],{"type":36},{"text":187,"type":33}," just to keep the models moving.",{"type":26,"content":189},[190],{"type":29,"content":191},[192,194],{"text":193,"type":33},"The real bottleneck isn't your model, ",{"text":195,"type":33,"marks":196},"it's the data.",[197],{"type":36},{"type":26,"content":199},[200],{"type":29,"content":201},[202,206,208],{"text":203,"type":33,"marks":204},"Better data",[205],{"type":36},{"text":207,"type":33}," is the last step before your ",{"text":209,"type":33,"marks":210},"robots are ready.",[211],{"type":36},"\u003C!--#storyblok#{\"name\": \"pain_points\", \"space\": \"290758475334152\", \"uid\": \"pain-001\", \"id\": \"148535978303585\"}-->","pain_points",{"type":26,"content":215},[216],{"type":29,"attrs":217,"content":219},{"textAlign":218},null,[220,222,226,228,232],{"text":221,"type":33},"You already have strong models, hardware, and a clear goal. What's missing is ",{"text":223,"type":33,"marks":224},"enough of the right data.",[225],{"type":36},{"text":227,"type":33}," Most teams don't have the setup to keep sending fresh, high-quality data to their models. And generating it takes ",{"text":229,"type":33,"marks":230},"time, people, and money. ",[231],{"type":36},{"text":233,"type":33},"That's why training slows down, tests fail, and launches move later and later on the calendar.",{"_uid":235,"stats":236,"visible":17,"_editable":253,"component":254,"transition_text":255,"transition_text_bold":256},"stats-001",[237,243,248],{"_uid":238,"label":239,"value":240,"_editable":241,"component":242},"st-001","Hours generated","+5000","\u003C!--#storyblok#{\"name\": \"stat_item\", \"space\": \"290758475334152\", \"uid\": \"st-001\", \"id\": \"148535978303585\"}-->","stat_item",{"_uid":244,"label":245,"value":246,"_editable":247,"component":242},"st-002","gbs Speed upload","+50","\u003C!--#storyblok#{\"name\": \"stat_item\", \"space\": \"290758475334152\", \"uid\": \"st-002\", \"id\": \"148535978303585\"}-->",{"_uid":249,"label":250,"value":251,"_editable":252,"component":242},"st-003","Time saving in data collection","60%","\u003C!--#storyblok#{\"name\": \"stat_item\", \"space\": \"290758475334152\", \"uid\": \"st-003\", \"id\": \"148535978303585\"}-->","\u003C!--#storyblok#{\"name\": \"stats\", \"space\": \"290758475334152\", \"uid\": \"stats-001\", \"id\": \"148535978303585\"}-->","stats","We take care of your data pipeline from ","start to finish.",{"_uid":258,"cards":259,"visible":17,"_editable":404,"component":405,"intro_text":55,"headline_bold":406,"headline_light":407,"intro_text_bold":55},"tw-001",[260,337],{"_uid":261,"title":262,"cta_link":263,"cta_text":264,"features":265,"subtitle":323,"_editable":335,"component":336},"twc-001","Off-the-shelf datasets",{"linktype":19,"cached_url":20},"Browse Datasets",[266,280,295,310],{"_uid":267,"text":268,"_editable":278,"component":279},"f-001",{"type":26,"content":269},[270],{"type":29,"content":271},[272,274],{"text":273,"type":33},"Instant access to ",{"text":275,"type":33,"marks":276},"ready-made robotics datasets.",[277],{"type":36},"\u003C!--#storyblok#{\"name\": \"feature_item\", \"space\": \"290758475334152\", \"uid\": \"f-001\", \"id\": \"148535978303585\"}-->","feature_item",{"_uid":281,"text":282,"_editable":294,"component":279},"f-002",{"type":26,"content":283},[284],{"type":29,"content":285},[286,288,292],{"text":287,"type":33},"Search by ",{"text":289,"type":33,"marks":290},"task, embodiment, DOFs, sensors, and environment",[291],{"type":36},{"text":293,"type":33}," and more.","\u003C!--#storyblok#{\"name\": \"feature_item\", \"space\": \"290758475334152\", \"uid\": \"f-002\", \"id\": \"148535978303585\"}-->",{"_uid":296,"text":297,"_editable":309,"component":279},"f-003",{"type":26,"content":298},[299],{"type":29,"content":300},[301,303,307],{"text":302,"type":33},"Clear ",{"text":304,"type":33,"marks":305},"pricing by hours",[306],{"type":36},{"text":308,"type":33}," of data or number of trajectories.","\u003C!--#storyblok#{\"name\": \"feature_item\", \"space\": \"290758475334152\", \"uid\": \"f-003\", \"id\": \"148535978303585\"}-->",{"_uid":311,"text":312,"_editable":322,"component":279},"f-004",{"type":26,"content":313},[314],{"type":29,"content":315},[316,320],{"text":317,"type":33,"marks":318},"New datasets added",[319],{"type":36},{"text":321,"type":33}," and improved over time as we collect more data.","\u003C!--#storyblok#{\"name\": \"feature_item\", \"space\": \"290758475334152\", \"uid\": \"f-004\", \"id\": \"148535978303585\"}-->",{"type":26,"content":324},[325],{"type":29,"content":326},[327,329,333],{"text":328,"type":33},"Perfect when you want to ",{"text":330,"type":33,"marks":331},"move now",[332],{"type":36},{"text":334,"type":33},": grab solid datasets and start training today.","\u003C!--#storyblok#{\"name\": \"two_ways_card\", \"space\": \"290758475334152\", \"uid\": \"twc-001\", \"id\": \"148535978303585\"}-->","two_ways_card",{"_uid":338,"title":339,"cta_link":340,"cta_text":341,"features":342,"subtitle":391,"_editable":403,"component":336},"twc-002","Bespoke data collection",{"linktype":19,"cached_url":53},"Request Bespoke Data",[343,356,369,378],{"_uid":344,"text":345,"_editable":355,"component":279},"f-005",{"type":26,"content":346},[347],{"type":29,"content":348},[349,353],{"text":350,"type":33,"marks":351},"Fully custom data",[352],{"type":36},{"text":354,"type":33}," collection built around your model, tasks, and goals.","\u003C!--#storyblok#{\"name\": \"feature_item\", \"space\": \"290758475334152\", \"uid\": \"f-005\", \"id\": \"148535978303585\"}-->",{"_uid":357,"text":358,"_editable":368,"component":279},"f-006",{"type":26,"content":359},[360],{"type":29,"content":361},[362,366],{"text":363,"type":33,"marks":364},"Custom",[365],{"type":36},{"text":367,"type":33}," embodiments, DOFs, sensors and environments.","\u003C!--#storyblok#{\"name\": \"feature_item\", \"space\": \"290758475334152\", \"uid\": \"f-006\", \"id\": \"148535978303585\"}-->",{"_uid":370,"text":371,"_editable":377,"component":279},"f-007",{"type":26,"content":372},[373],{"type":29,"content":374},[375],{"text":376,"type":33},"Ongoing collects of new data that match your training cycle and releases.","\u003C!--#storyblok#{\"name\": \"feature_item\", \"space\": \"290758475334152\", \"uid\": \"f-007\", \"id\": \"148535978303585\"}-->",{"_uid":379,"text":380,"_editable":390,"component":279},"f-008",{"type":26,"content":381},[382],{"type":29,"content":383},[384,386],{"text":385,"type":33},"Designed around your training ",{"text":387,"type":33,"marks":388},"objectives and KPIs.",[389],{"type":36},"\u003C!--#storyblok#{\"name\": \"feature_item\", \"space\": \"290758475334152\", \"uid\": \"f-008\", \"id\": \"148535978303585\"}-->",{"type":26,"content":392},[393],{"type":29,"content":394},[395,397,401],{"text":396,"type":33},"Ideal if you're ",{"text":398,"type":33,"marks":399},"ready to build serious",[400],{"type":36},{"text":402,"type":33}," robots and need data that fits your setup and needs.","\u003C!--#storyblok#{\"name\": \"two_ways_card\", \"space\": \"290758475334152\", \"uid\": \"twc-002\", \"id\": \"148535978303585\"}-->","\u003C!--#storyblok#{\"name\": \"two_ways\", \"space\": \"290758475334152\", \"uid\": \"tw-001\", \"id\": \"148535978303585\"}-->","two_ways","the data you need","Two ways to get exactly",{"_uid":409,"cards":410,"image":452,"visible":17,"_editable":457,"component":458,"headline_bold":459,"headline_light":460},"why-001",[411,422,432,442],{"_uid":412,"title":413,"_editable":414,"component":415,"description":416},"wc-001","We're your neutral data partner","\u003C!--#storyblok#{\"name\": \"why_card\", \"space\": \"290758475334152\", \"uid\": \"wc-001\", \"id\": \"148535978303585\"}-->","why_card",{"type":26,"content":417},[418],{"type":29,"content":419},[420],{"text":421,"type":33},"We work for your product. Our goal is to help you train better robots. You keep full control of your roadmap while we focus on supplying the right data behind it.",{"_uid":423,"title":424,"_editable":425,"component":415,"description":426},"wc-002","We only do robotics data","\u003C!--#storyblok#{\"name\": \"why_card\", \"space\": \"290758475334152\", \"uid\": \"wc-002\", \"id\": \"148535978303585\"}-->",{"type":26,"content":427},[428],{"type":29,"content":429},[430],{"text":431,"type":33},"We're not a model shop or a generic AI agency. Our work is to design, collect, and deliver the datasets your team needs so you can spend your time on training and building.",{"_uid":433,"title":434,"_editable":435,"component":415,"description":436},"wc-003","Built by experts","\u003C!--#storyblok#{\"name\": \"why_card\", \"space\": \"290758475334152\", \"uid\": \"wc-003\", \"id\": \"148535978303585\"}-->",{"type":26,"content":437},[438],{"type":29,"content":439},[440],{"text":441,"type":33},"Our team is advised by experts in the field. We speak your language and shape each project around real research and product needs.",{"_uid":443,"title":444,"_editable":445,"component":415,"description":446},"wc-004","Infrastructure and management already in place","\u003C!--#storyblok#{\"name\": \"why_card\", \"space\": \"290758475334152\", \"uid\": \"wc-004\", \"id\": \"148535978303585\"}-->",{"type":26,"content":447},[448],{"type":29,"content":449},[450],{"text":451,"type":33},"We bring the lab, hardware, operators, and management needed to run a serious collection. You don't have to hire a new team or set up new facilities just to get the datasets you need.",{"id":218,"alt":453,"name":55,"focus":55,"title":55,"source":55,"filename":454,"copyright":55,"fieldtype":455,"meta_data":456,"is_external_url":92},"Robot interacting with colorful toys","https:\u002F\u002Fa.storyblok.com\u002Ff\u002F290758475334152\u002F159000\u002Fb05c6ebc44\u002Fwhy-choose-robot.png","asset",{},"\u003C!--#storyblok#{\"name\": \"why_choose\", \"space\": \"290758475334152\", \"uid\": \"why-001\", \"id\": \"148535978303585\"}-->","why_choose","Why choose us","for your robotics data?",{"_uid":462,"subtext":463,"visible":17,"cta_link":469,"cta_text":470,"headline":471,"_editable":477,"component":478,"show_waves":17,"cta_secondary_link":479,"cta_secondary_text":480,"cta_secondary_color":55},"ctaf-001",{"type":26,"content":464},[465],{"type":29,"content":466},[467],{"text":468,"type":33},"We'll guide you through every step.",{"linktype":19,"cached_url":20},"Browse datasets",{"type":26,"content":472},[473],{"type":29,"content":474},[475],{"text":476,"type":33},"Ready to unlock better robotics training?","\u003C!--#storyblok#{\"name\": \"cta_footer\", \"space\": \"290758475334152\", \"uid\": \"ctaf-001\", \"id\": \"148535978303585\"}-->","cta_footer",{"linktype":19,"cached_url":53},"Request bespoke data",{"_uid":482,"year":483,"email":484,"address":485,"visible":17,"_editable":486,"component":487,"copyright":488,"brand_name":489,"linkedin_url":490},"footer-001","2025","contact@nurvai.ai","Austin 305 W 18th St, TX 78701, USA","\u003C!--#storyblok#{\"name\": \"site_footer\", \"space\": \"290758475334152\", \"uid\": \"footer-001\", \"id\": \"148535978303585\"}-->","site_footer","All Rights Reserved","nurvai","https:\u002F\u002Fwww.linkedin.com\u002Fcompany\u002Fnurvai","\u003C!--#storyblok#{\"name\": \"page\", \"space\": \"290758475334152\", \"uid\": \"0285b3db-5933-4051-967a-420df86db660\", \"id\": \"148535978303585\"}-->","page","NurvAI — Human-Created Datasets for Robotics","Nurvai delivers human-created robotics datasets, collected through teleoperation, egocentric data capture, and sensors for real-world training and deployment.","home",-40,[],0,"2c8aa1f6-ac1b-4ccd-b565-8e6349747fc5","2026-02-24T19:02:43.271Z","default",[],1776880649,[],[],{"age":507,"cache-control":508,"connection":509,"content-encoding":510,"content-type":511,"date":512,"etag":513,"referrer-policy":514,"sb-be-version":515,"server":516,"transfer-encoding":517,"vary":518,"via":519,"x-amz-cf-id":520,"x-amz-cf-pop":521,"x-cache":522,"x-content-type-options":523,"x-frame-options":524,"x-permitted-cross-domain-policies":525,"x-request-id":526,"x-runtime":527,"x-xss-protection":528},"12206","max-age=0, public, s-maxage=604800, stale-if-error=3600","keep-alive","gzip","application\u002Fjson; charset=utf-8","Wed, 22 Apr 2026 17:58:31 GMT","W\u002F\"3a94d10db9f92494d084d2d1dd72e486\"","strict-origin-when-cross-origin","5.764.0","nginx\u002F1.29.1","chunked","Origin,Accept-Encoding","1.1 9b54c47ba82fa634dd5f3b43f670c934.cloudfront.net (CloudFront)","mrfnXN3KcRt2aQRfwiab4Y1RPvGshp2c5LXGhKV8v0dUccV9fUaTeg==","FRA60-P12","Hit from cloudfront","nosniff","SAMEORIGIN","none","a1240fea-4baa-4e17-8d4d-5df0181a2181","0.027344","0",{"data":530,"headers":599},{"story":531,"cv":503,"rels":597,"links":598},{"name":532,"created_at":533,"published_at":534,"updated_at":535,"id":536,"uuid":537,"content":538,"slug":590,"full_slug":591,"sort_by_date":218,"position":592,"tag_list":593,"is_startpage":17,"parent_id":594,"meta_data":218,"group_id":595,"first_published_at":534,"release_id":218,"lang":501,"path":218,"alternates":596,"default_full_slug":218,"translated_slugs":218},"Journal Index","2026-04-09T20:45:39.981Z","2026-04-09T20:45:40.027Z","2026-04-09T20:45:40.045Z",164159242192390,"cc5684a5-7e7f-4d44-beea-0a109ad4d2d4",{"_uid":539,"body":540,"component":492,"seo_title":588,"seo_description":589},"622b0ebd",[541,565,573],{"_uid":542,"links":543,"visible":17,"cta_link":563,"cta_text":55,"component":95},"0ad927b2",[544,547,550,554,557,560],{"_uid":545,"link":546,"label":6,"component":64},"407a59cf",{"linktype":19,"cached_url":62},{"_uid":548,"link":549,"label":68,"component":64,"coming_soon":17},"a457dbcf",{"linktype":19,"cached_url":20},{"_uid":551,"link":552,"label":553,"component":64},"bb347a24",{"linktype":19,"cached_url":53},"Bespoke Data",{"_uid":555,"link":556,"label":79,"component":64},"bd5aaa36",{"linktype":19,"cached_url":78},{"_uid":558,"link":559,"label":85,"component":64},"125bde3c",{"linktype":19,"cached_url":84},{"_uid":561,"link":562,"label":91,"component":64,"coming_soon":92},"1eb5128e",{"linktype":19,"cached_url":90},{"linktype":19,"cached_url":564},"\u002Flogin",{"_uid":566,"visible":17,"headline":567,"all_label":568,"component":569,"hero_label":91,"posts_per_page":570,"headline_accent":571,"load_more_label":572},"3f90590c","& Research","All","blog_index",10,"Insights","Load more",{"_uid":574,"subtext":575,"visible":17,"cta_link":580,"cta_soon":17,"cta_text":470,"headline":581,"component":478,"cta_color":24,"cta_active":92,"cta_secondary_link":586,"cta_secondary_soon":17,"cta_secondary_text":480,"cta_secondary_color":587,"cta_secondary_active":92},"c889a74e",{"type":26,"content":576},[577],{"type":29,"content":578},[579],{"text":468,"type":33},{"linktype":19,"cached_url":20},{"type":26,"content":582},[583],{"type":29,"content":584},[585],{"text":476,"type":33},{"linktype":19,"cached_url":53},"indigo","Journal | NurvAI","Insights and research from NurvAI on robotics data collection, teleoperation, and training dataset quality.","journal","journal\u002F",-120,[],164159239587333,"af32e251-1fec-47eb-87eb-8cb22cfa3f21",[],[],[],{"age":600,"cache-control":508,"connection":509,"content-encoding":510,"content-type":511,"date":601,"etag":602,"referrer-policy":514,"sb-be-version":515,"server":516,"transfer-encoding":517,"vary":518,"via":519,"x-amz-cf-id":603,"x-amz-cf-pop":521,"x-cache":522,"x-content-type-options":523,"x-frame-options":524,"x-permitted-cross-domain-policies":525,"x-request-id":604,"x-runtime":605,"x-xss-protection":528},"12091","Wed, 22 Apr 2026 18:00:26 GMT","W\u002F\"f5a6f87ff9b7f0a600a87c6fae9082b7\"","2Yjau7jnid7DmJgfORtf11xI8FbtgzAeWUEVclLkc2cUfrv_CD-_9g==","eb27cd93-81e8-4db0-a3e9-9532c87cf22d","0.026317",{"_uid":607,"component":608,"coming_soon_enabled":92,"coming_soon_subtext":609,"coming_soon_headline":610},"276b5df2","site_settings","Join the waitlist. Be first to know when we launch.","Ready to \u003Cstrong>scale your robotics\u003C\u002Fstrong> with human-created datasets?",[612],{"name":613,"created_at":614,"published_at":615,"updated_at":616,"id":617,"uuid":618,"content":619,"slug":952,"full_slug":953,"sort_by_date":218,"position":954,"tag_list":955,"is_startpage":92,"parent_id":594,"meta_data":218,"group_id":956,"first_published_at":957,"release_id":218,"lang":501,"path":218,"alternates":958,"default_full_slug":218,"translated_slugs":218},"Why Real-World Data is Relevant for Robotics Training","2026-04-22T16:06:02.598Z","2026-04-22T16:10:36.944Z","2026-04-22T16:10:36.959Z",168691149334456,"3454f41d-bb14-4ad9-aeb8-8c97904f62fa",{"_uid":620,"body":621,"tags":942,"title":613,"excerpt":943,"category":944,"component":945,"seo_title":946,"author_name":947,"author_role":944,"reading_time":948,"featured_image":949,"seo_description":943},"1affc17c-aa97-4c49-80f1-cbba2280ab55",{"type":26,"content":622},[623,627,631,635,642,646,652,656,661,665,670,674,679,683,687,692,696,701,705,709,713,718,722,726,730,734,739,743,747,751,755,760,764,769,778,783,793,797,802,806,810,814,818,831,836],{"type":29,"content":624},[625],{"text":626,"type":33},"As embodied AI continues to advance, data is quickly becoming the central bottleneck in developing new models. Robot data is expensive to collect, difficult to obtain, and inherently hard to scale [1]; [2]. Yet despite these challenges, real-world data continues to drive many of the most robust and deployable robotics systems. To address this constraint, the industry has increasingly turned to alternatives such as world models, simulation environments, and generative video models to help meet the growing demand for training data [3].",{"type":29,"content":628},[629],{"text":630,"type":33},"But while these approaches are promising, an important question remains: can simulated data truly replace real-world experience?",{"type":29,"content":632},[633],{"text":634,"type":33},"We argue that real-world robotic data, despite its cost and complexity, ultimately creates the most durable and defensible moat in embodied AI.",{"type":636,"attrs":637,"content":639},"heading",{"level":638},2,[640],{"text":641,"type":33},"Why is Data Collection so Difficult?",{"type":29,"content":643},[644],{"text":645,"type":33},"Collecting real-world robot data requires significant infrastructure and coordination. Unlike traditional machine learning datasets, robotics data collection is an operational process involving hardware, people, and physical environments [1].",{"type":636,"attrs":647,"content":649},{"level":648},3,[650],{"text":651,"type":33},"Hardware Requirements",{"type":29,"content":653},[654],{"text":655,"type":33},"Data collection requires specialized hardware and capture mechanisms. These vary widely in cost depending on the type of data being collected. For example, egocentric data collection can be relatively inexpensive, sometimes requiring only a single camera or smartphone. In contrast, teleoperation data is significantly more costly, as even the simplest setup requires a robot embodiment, control interfaces, and safety infrastructure.",{"type":636,"attrs":657,"content":658},{"level":648},[659],{"text":660,"type":33},"Trained Operators",{"type":29,"content":662},[663],{"text":664,"type":33},"Data collection requires a team of trained operators. These operators must be capable of performing tasks consistently while generating high-quality demonstrations. Training operators and maintaining productivity adds both time and financial costs to the data collection process.",{"type":636,"attrs":666,"content":667},{"level":648},[668],{"text":669,"type":33},"Task Design",{"type":29,"content":671},[672],{"text":673,"type":33},"Task design plays a critical role. Task designers must understand industry needs, define meaningful scenarios, and ensure that demonstrations capture useful behaviors. Without careful task design, large amounts of collected data may have limited training value.",{"type":636,"attrs":675,"content":676},{"level":648},[677],{"text":678,"type":33},"Infrastructure and Data Pipelines",{"type":29,"content":680},[681],{"text":682,"type":33},"Real-world data collection requires both physical and digital infrastructure. Teams need dedicated environments for data capture, including workspace, objects, and safety setups. In addition, large-scale robotics datasets require substantial storage, processing, and data management pipelines.",{"type":29,"content":684},[685],{"text":686,"type":33},"Together, these requirements make real-world robotics data collection both costly and difficult to scale. As a result, optimizing data collection workflows becomes a key challenge for organizations building robotics datasets.",{"type":636,"attrs":688,"content":689},{"level":638},[690],{"text":691,"type":33},"Why Real-World Data Remains Critical",{"type":29,"content":693},[694],{"text":695,"type":33},"Despite the difficulties in collecting it, real-world data presents several advantages over simulated data. These advantages primarily relate to robustness, generalization, and deployment reliability in unconstrained environments. The following sections outline key limitations of simulated data and why real-world data remains critical for building deployable robotic systems.",{"type":636,"attrs":697,"content":698},{"level":648},[699],{"text":700,"type":33},"The Sim-To-Real Gap",{"type":29,"content":702},[703],{"text":704,"type":33},"The sim-to-real gap refers to the difficulty of transferring policies trained in simulation to real-world environments. This gap exists because simulated physics engines fail to fully capture the stochasticity and complex dynamics of the physical world. These limitations become especially noticeable in tasks involving cloth and deformable objects. Soft-body interactions, contact dynamics, and fine-grained material behavior remain difficult to simulate accurately, often due to constraints in mesh resolution and computational cost. As a result, models trained purely in simulation can perform well in controlled virtual environments but struggle when exposed to real-world variability [4]; [3].",{"type":29,"content":706},[707],{"text":708,"type":33},"Real-world data helps bypass this issue by cutting out the middleman. Instead of learning from approximations, models are trained directly on the environments in which they will ultimately operate.",{"type":29,"content":710},[711],{"text":712,"type":33},"While recent advances in simulation, domain randomization, and world models have helped narrow this gap, sim-to-real transfer remains one of the biggest challenges in robotics training today [3].",{"type":636,"attrs":714,"content":715},{"level":648},[716],{"text":717,"type":33},"Generalization Challenges",{"type":29,"content":719},[720],{"text":721,"type":33},"Simulated environments often struggle to capture many of the complexities present in real-world settings, which makes generalization difficult. In real environments, lighting conditions change, humans move unpredictably around the robot, objects become cluttered, and occlusions frequently occur. These variations introduce a level of uncertainty that is difficult to fully replicate in simulation.",{"type":29,"content":723},[724],{"text":725,"type":33},"Even when simulators attempt to introduce variability through domain randomization, they typically fail to capture the full diversity and long-tail distribution of real-world scenarios. Small differences, such as sensor noise, object wear, reflections, or unexpected interactions, can significantly impact robot performance.",{"type":29,"content":727},[728],{"text":729,"type":33},"As a result, models trained primarily in simulation often perform well in controlled environments but struggle when deployed in dynamic, unstructured settings.",{"type":29,"content":731},[732],{"text":733,"type":33},"Real-world data helps address this challenge by exposing models to natural variability from the start, improving robustness and enabling better generalization across environments [2]; [5].",{"type":636,"attrs":735,"content":736},{"level":648},[737],{"text":738,"type":33},"Reliability and Safety",{"type":29,"content":740},[741],{"text":742,"type":33},"Robots deployed in real-world environments must meet a high bar for reliability and safety. Unlike purely digital systems, robots interact directly with people, objects, and infrastructure, meaning failures can result in physical damage, safety risks, or costly downtime. As a result, robotic systems must learn to perform actions in ways that avoid both self-damage and unintended impacts on their surroundings.",{"type":29,"content":744},[745],{"text":746,"type":33},"Simulation plays an important role early in development. It allows teams to safely test behaviors, explore edge cases, and iterate quickly without risking hardware or environments. However, as systems move closer to deployment, training exclusively in simulated environments becomes less effective. Subtle factors such as sensor noise, wear and tear, latency, calibration drift, and unpredictable human interactions are difficult to model accurately, yet they are critical for ensuring reliable performance.",{"type":29,"content":748},[749],{"text":750,"type":33},"Because of this, achieving production-level reliability ultimately depends on exposure to real-world conditions [1]; [5]. Real-world data enables models to learn from true operational variability, failure cases, and edge scenarios that simulations often miss.",{"type":29,"content":752},[753],{"text":754,"type":33},"As robotics transitions from controlled demos to real deployments, real-world data becomes essential not just for performance, but for building systems that are safe, robust, and ready for everyday use [6].",{"type":636,"attrs":756,"content":757},{"level":638},[758],{"text":759,"type":33},"Real-World Success Cases",{"type":29,"content":761},[762],{"text":763,"type":33},"Recent progress in embodied AI further reinforces the importance of real-world data. Several of the most capable robotic systems today rely heavily on real-world interaction and large-scale physical datasets.",{"type":636,"attrs":765,"content":766},{"level":648},[767],{"text":768,"type":33},"π*₀.₆",{"type":29,"content":770},[771,773,776],{"text":772,"type":33},"Physical Intelligence's ",{"text":768,"type":33,"marks":774},[775],{"type":36},{"text":777,"type":33}," is one such example. The model improves through real-world interaction, incorporating reinforcement learning from deployment experience in addition to demonstration data. By learning from real-world successes, failures, and corrective interventions, the system is able to improve robustness and performance on long-horizon manipulation tasks.",{"type":636,"attrs":779,"content":780},{"level":648},[781],{"text":782,"type":33},"GEN-1",{"type":29,"content":784},[785,787,791],{"text":786,"type":33},"Similarly, Generalist's ",{"text":788,"type":33,"marks":789},"Gen-1",[790],{"type":36},{"text":792,"type":33}," model was pretrained on large-scale real-world manipulation datasets, including UMI-style data. These datasets emphasize diverse demonstrations collected across tasks, environments, and embodiments. Pretraining on this type of real-world data enables models to develop broader capabilities and adapt more effectively to new scenarios.",{"type":29,"content":794},[795],{"text":796,"type":33},"These examples highlight a broader trend in robotics: while simulation and synthetic data continue to improve, many of the most capable systems still depend heavily on real-world data for robustness, reliability, and deployment readiness.",{"type":636,"attrs":798,"content":799},{"level":638},[800],{"text":801,"type":33},"The Strategic Importance of Real-World Data",{"type":29,"content":803},[804],{"text":805,"type":33},"As embodied AI continues to evolve, data is emerging as one of the most important drivers of progress. While simulation, world models, and generative approaches offer promising paths to scale, they remain limited by their ability to fully capture the complexity of the physical world.",{"type":29,"content":807},[808],{"text":809,"type":33},"Real-world data, despite being difficult and expensive to collect, provides advantages that are difficult to replicate synthetically. It helps close the sim-to-real gap, improves generalization to dynamic environments, and enables the level of reliability required for real-world deployment.",{"type":29,"content":811},[812],{"text":813,"type":33},"For organizations building embodied AI systems, this creates a clear strategic implication. The ability to collect, curate, and scale real-world robotics data is not just an operational challenge, but a long-term competitive advantage.",{"type":29,"content":815},[816],{"text":817,"type":33},"As robotics transitions from research prototypes to production systems, real-world data is likely to remain a critical foundation for building robust, deployable, and general-purpose robotic intelligence.",{"type":29,"content":819},[820,822,829],{"text":821,"type":33},"At Nurvai, we're working to make real-world robotics data easier to collect, scale, and deploy. If you're building embodied AI systems and looking to accelerate development, ",{"text":823,"type":33,"marks":824},"get in touch",[825],{"type":826,"attrs":827},"link",{"href":53,"uuid":218,"anchor":218,"target":828,"linktype":19},"_blank",{"text":830,"type":33},".",{"type":636,"attrs":832,"content":833},{"level":638},[834],{"text":835,"type":33},"References",{"type":837,"content":838},"bullet_list",[839,857,873,889,911,927],{"type":840,"content":841},"list_item",[842],{"type":29,"content":843},[844,848,850,855],{"text":845,"type":33,"marks":846},"[1] ",[847],{"type":36},{"text":849,"type":33},"D. Kalashnikov, A. Irpan, P. Pastor, et al., \"QT-Opt: Scalable deep reinforcement learning for vision-based robotic manipulation,\" ",{"text":851,"type":33,"marks":852},"arXiv preprint arXiv:1806.10293",[853],{"type":854},"italic",{"text":856,"type":33},", 2018.",{"type":840,"content":858},[859],{"type":29,"content":860},[861,865,867,871],{"text":862,"type":33,"marks":863},"[2] ",[864],{"type":36},{"text":866,"type":33},"O. X.-E. Collaboration, \"Open X-Embodiment: Robotic learning datasets and RT-X models,\" ",{"text":868,"type":33,"marks":869},"arXiv preprint arXiv:2310.08864",[870],{"type":854},{"text":872,"type":33},", 2023.",{"type":840,"content":874},[875],{"type":29,"content":876},[877,881,883,887],{"text":878,"type":33,"marks":879},"[3] ",[880],{"type":36},{"text":882,"type":33},"F. Muratore, M. Gienger, J. Peters, \"A survey on sim-to-real transfer for robotics,\" ",{"text":884,"type":33,"marks":885},"Foundations and Trends in Robotics",[886],{"type":854},{"text":888,"type":33},", vol. 9, no. 1–2, pp. 1–142, 2022. doi: 10.1561\u002F2300000079.",{"type":840,"content":890},[891],{"type":29,"content":892},[893,897,899,903,905],{"text":894,"type":33,"marks":895},"[4] ",[896],{"type":36},{"text":898,"type":33},"J. Tobin, R. Fong, A. Ray, et al., \"Domain randomization for transferring deep neural networks from simulation to the real world,\" ",{"text":900,"type":33,"marks":901},"IEEE\u002FRSJ International Conference on Intelligent Robots and Systems (IROS)",[902],{"type":854},{"text":904,"type":33},", 2017. ",{"text":906,"type":33,"marks":907},"arxiv.org\u002Fabs\u002F1703.06907",[908],{"type":826,"attrs":909},{"href":910,"uuid":218,"anchor":218,"target":828,"linktype":19},"https:\u002F\u002Farxiv.org\u002Fabs\u002F1703.06907",{"type":840,"content":912},[913],{"type":29,"content":914},[915,919,921,925],{"text":916,"type":33,"marks":917},"[5] ",[918],{"type":36},{"text":920,"type":33},"A. Brohan, N. Brown, J. Carbajal, Y. Chebotar, et al., \"RT-1: Robotics transformer for real-world control at scale,\" ",{"text":922,"type":33,"marks":923},"arXiv preprint arXiv:2212.06817",[924],{"type":854},{"text":926,"type":33},", 2022.",{"type":840,"content":928},[929],{"type":29,"content":930},[931,935,937,941],{"text":932,"type":33,"marks":933},"[6] ",[934],{"type":36},{"text":936,"type":33},"A. Brohan, Y. Chebotar, C. Finn, S. Levine, et al., \"RT-2: Vision-language-action models transfer web knowledge to robotic control,\" ",{"text":938,"type":33,"marks":939},"arXiv preprint arXiv:2307.15818",[940],{"type":854},{"text":872,"type":33},"Real-World Data, Sim-to-Real, Embodied AI, Training","Real-world robotic data is expensive and operationally hard to scale, yet simulation still can't replace it. Why real-world data is the most durable moat in embodied AI.","Research","blog_post","Why Real-World Data is Relevant for Robotics Training | NurvAI","Mateo Tanaka","10 min read",{"id":950,"alt":613,"filename":951},168690111491565,"https:\u002F\u002Fa.storyblok.com\u002Ff\u002F290758475334152\u002F3731507\u002F57255121bb\u002Fclaudio-schwarz-journal-cover.jpg","real-world-data-robotics-training","journal\u002Freal-world-data-robotics-training",-110,[],"7bf9c6d3-a693-42ab-b557-36521bd3079d","2026-04-22T16:06:02.744Z",[]]