--- title: MolmoMotion: Language-guided 3D motion forecasting description: MolmoMotion is a new language-guided 3D motion forecasting model from AI2 that predicts where marked 3D points on objects will move over the next few seconds,... canonical: https://daily.dev/posts/molmomotion-language-guided-3d-motion-forecasting-j9ihm74ru twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: MolmoMotion: Language-guided 3D motion forecasting | daily.dev og:description: MolmoMotion is a new language-guided 3D motion forecasting model from AI2 that predicts where marked 3D points on objects will move over the next few seconds,... og:url: https://daily.dev/posts/molmomotion-language-guided-3d-motion-forecasting-j9ihm74ru og:image: https://api.daily.dev/og/posts/J9IHm74Ru.png og:image:alt: MolmoMotion: Language-guided 3D motion forecasting og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # MolmoMotion: Language-guided 3D motion forecasting **[Hugging Face](https://daily.dev/sources/huggingface)** · 9 min read · 0 upvotes · 0 comments ## Summary MolmoMotion is a new language-guided 3D motion forecasting model from AI2 that predicts where marked 3D points on objects will move over the next few seconds, given a video frame and a natural language action description. It uses Molmo 2 as its backbone and comes in two variants: an autoregressive version (MolmoMotion-AR) for well-defined trajectories and a flow-matching version (MolmoMotion-FM) for uncertain multi-modal futures. Alongside the model, AI2 releases MolmoMotion-1M, a dataset of 1.16M videos with object-grounded 3D point trajectories paired with action descriptions, and PointMotionBench, a 2.7K-clip human-validated benchmark. MolmoMotion outperforms existing 3D forecasting methods on the benchmark, improves robot pick-and-place success from 56% to 76.3%, and enhances motion quality in video generation tasks. Model weights, dataset, and benchmark are all openly released. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What is MolmoMotion and what does it predict? MolmoMotion is a language-guided 3D motion forecasting model from Ai2 that predicts how 3D points marked on an object will move over the next few seconds, given a video frame, query points, and a written action description like 'move and rotate the wooden bowl.' It uses the Molmo 2 backbone and comes in autoregressive (MolmoMotion-AR) and flow-matching (MolmoMotion-FM) variants. _Teams evaluating motion-forecasting approaches for robotics or video pipelines can track releases like MolmoMotion on daily.dev._ ### How much does MolmoMotion improve robot pick-and-place success rates over a Molmo 2 baseline? A control policy built on MolmoMotion succeeds on 76.3% of pick-and-place tasks in simulation, compared to 56.0% for the same policy built on plain Molmo 2. It also learns faster, reaching 51% success after 10K training steps versus Molmo 2's 19%, and on real robots it matches Molmo 2's 12K-step test error in only about 2K training steps. _Robotics engineers comparing motion-forecasting backbones can follow benchmark results like these on daily.dev._ ### What datasets did Ai2 release alongside MolmoMotion? Ai2 released MolmoMotion-1M, drawn from 1.16 million videos and spanning 736 motion types and 5.6K distinct objects, described as the largest corpus of action-described, object-grounded 3D point trajectories to date. It also released PointMotionBench, a human-validated benchmark of 2.7K clips across 111 object categories and 61 motion types for evaluating 3D motion forecasting accuracy. _Researchers building or benchmarking motion-forecasting models can keep tabs on new open datasets via daily.dev._ ## Similar posts on daily.dev - [The $250 Million Paper](https://daily.dev/posts/the-250-million-paper-ciwojhnk2) · ByteByteGo · 2 upvotes · 0 comments - [Ai2 launches MolmoWeb, an open-source web agent](https://daily.dev/posts/ai2-launches-molmoweb-an-open-source-web-agent-ec7sqrixm) · The New Stack · 4 upvotes · 0 comments - [FLUX 3 x mimic: The Next Generation of Video-Action Models](https://daily.dev/posts/flux-3-x-mimic-the-next-generation-of-video-action-models-ixm7y902g) · Hacker News · 0 upvotes · 0 comments --- Tags: [#deep-learning](https://daily.dev/tags/deep-learning), [#computer-vision](https://daily.dev/tags/computer-vision), [#robotics](https://daily.dev/tags/robotics), [#video-generation](https://daily.dev/tags/video-generation) [View this post on daily.dev](https://daily.dev/posts/molmomotion-language-guided-3d-motion-forecasting-j9ihm74ru) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"MolmoMotion: Language-guided 3D motion forecasting","url":"https://daily.dev/posts/molmomotion-language-guided-3d-motion-forecasting-j9ihm74ru","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/molmomotion-language-guided-3d-motion-forecasting-j9ihm74ru"},"datePublished":"2026-06-17T15:29:55.860Z","dateModified":"2026-09-14T06:21:35.239Z","description":"MolmoMotion is a new language-guided 3D motion forecasting model from AI2 that predicts where marked 3D points on objects will move over the next few seconds,...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/2e3279fd2e53d6b328f5cc35821cde12?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/2e3279fd2e53d6b328f5cc35821cde12?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Hugging Face","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Hugging Face","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/f1f55c67d81a4330acf5b90b26b0c8e1","url":"https://daily.dev/sources/huggingface"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/molmomotion-language-guided-3d-motion-forecasting-j9ihm74ru","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"deep-learning,computer-vision,robotics,video-generation","timeRequired":"PT9M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Hugging Face","item":"https://daily.dev/sources/huggingface"},{"@type":"ListItem","position":3,"name":"MolmoMotion: Language-guided 3D motion forecasting"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/molmomotion-language-guided-3d-motion-forecasting-j9ihm74ru#faq","mainEntity":[{"@type":"Question","name":"What is MolmoMotion and what does it predict?","acceptedAnswer":{"@type":"Answer","text":"MolmoMotion is a language-guided 3D motion forecasting model from Ai2 that predicts how 3D points marked on an object will move over the next few seconds, given a video frame, query points, and a written action description like 'move and rotate the wooden bowl.' It uses the Molmo 2 backbone and comes in autoregressive (MolmoMotion-AR) and flow-matching (MolmoMotion-FM) variants. Teams evaluating motion-forecasting approaches for robotics or video pipelines can track releases like MolmoMotion on daily.dev."}},{"@type":"Question","name":"How much does MolmoMotion improve robot pick-and-place success rates over a Molmo 2 baseline?","acceptedAnswer":{"@type":"Answer","text":"A control policy built on MolmoMotion succeeds on 76.3% of pick-and-place tasks in simulation, compared to 56.0% for the same policy built on plain Molmo 2. It also learns faster, reaching 51% success after 10K training steps versus Molmo 2's 19%, and on real robots it matches Molmo 2's 12K-step test error in only about 2K training steps. Robotics engineers comparing motion-forecasting backbones can follow benchmark results like these on daily.dev."}},{"@type":"Question","name":"What datasets did Ai2 release alongside MolmoMotion?","acceptedAnswer":{"@type":"Answer","text":"Ai2 released MolmoMotion-1M, drawn from 1.16 million videos and spanning 736 motion types and 5.6K distinct objects, described as the largest corpus of action-described, object-grounded 3D point trajectories to date. It also released PointMotionBench, a human-validated benchmark of 2.7K clips across 111 object categories and 61 motion types for evaluating 3D motion forecasting accuracy. Researchers building or benchmarking motion-forecasting models can keep tabs on new open datasets via daily.dev."}}]} ```