{"posts":[{"id":20313,"title":"Turn Any Real-World Space Into a Robot Training Ground For Free","excerpt":"Today we are announcing a new capability that connects everything we have been building directly to the robotics world: free USDZ export. Starting now, anyone can capture a real-world space with a smartphone or a 360 camera, and download it as a metric-scaled, simulation-ready USDZ file, containing both a photorealistic 3D Gaussian splat and a [&hellip;]","content":"<p><span style=\"font-weight: 400;\">Today we are announcing a new capability that connects everything we have been building directly to the robotics world: <\/span><b>free USDZ export<\/b><span style=\"font-weight: 400;\">.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Starting now, anyone can capture a real-world space with a smartphone or a 360 camera, and download it as a <\/span><b>metric-scaled, simulation-ready USDZ file<\/b><span style=\"font-weight: 400;\">, containing both a photorealistic 3D Gaussian splat and a collision-ready mesh, ready to be imported into NVIDIA Isaac Sim.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">No LiDAR rig. No enterprise sales call. No waiting list. Capture, upload, download, simulate.<\/span><\/p>\n<p>&nbsp;<\/p>\n<h2><span style=\"font-weight: 400;\">One File, Two Representations, True Scale<\/span><\/h2>\n<p><span style=\"font-weight: 400;\">Every capture processed through OVER can now be exported as a single USDZ file that packages:<\/span><\/p>\n<ul>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>A high-fidelity 3D Gaussian splat<\/b><span style=\"font-weight: 400;\"> : the robot&#8217;s eyes. It renders photorealistic RGB from any viewpoint, reproducing the real lighting, textures, reflections, and clutter of the original space<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>A collision mesh derived directly from the splat<\/b><span style=\"font-weight: 400;\"> : the robot&#8217;s body. It gives the physics engine the geometry it needs for contact, navigation, and manipulation<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>True metric scale<\/b><span style=\"font-weight: 400;\"> : one meter in the real world is one meter in the simulation<\/span><\/li>\n<\/ul>\n<p><span style=\"font-weight: 400;\">Because the mesh is generated from the same reconstruction as the splat, the two representations are always aligned. There is no manual registration step, no scale guessing, and no drift between what the robot sees and what the physics engine computes. Import the file into Isaac Sim and the environment simply works.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">This pairing is what makes the asset unique. Inside a single environment, a robot can physically interact with the scene through the collision mesh while its perception stack is fed photorealistic RGB rendered from the splat, the same kind of visual input its onboard cameras will receive in the real world. That closes the loop for vision-based policies: act, collide, observe, repeat, with both the physics and the pixels grounded in the same real place.<\/span><\/p>\n<p>&nbsp;<\/p>\n<h2><span style=\"font-weight: 400;\">Why This Matters for Robotics<\/span><\/h2>\n<p><span style=\"font-weight: 400;\">Anyone training robots today knows the problem: the <\/span><b>sim-to-real gap<\/b><span style=\"font-weight: 400;\">. Policies trained in clean, synthetic environments consistently struggle when deployed in the messy real world, because synthetic scenes lack the visual richness reality provides, imperfect lighting, worn surfaces, reflections, and clutter.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">The best way to close that gap is to train in a faithful digital twin of the environment where the robot will actually operate. Until now, producing that twin meant expensive RGB-LiDAR rigs, specialized survey teams, and budgets that only large labs could justify.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">With free USDZ export, the capture device becomes the smartphone already in your pocket, or a consumer 360 camera. Walk through a warehouse, a retail floor, a street, or a lab. A few minutes later, that exact space is a training environment, with correct distances, correct proportions, and real-world visual complexity that synthetic assets can&#8217;t replicate.<\/span><\/p>\n<p>&nbsp;<\/p>\n<h2><span style=\"font-weight: 400;\">How It Works<\/span><\/h2>\n<p><b>1. Capture<\/b><span style=\"font-weight: 400;\"> the space with your smartphone or a 360 camera through the OVER app: <\/span><a href=\"https:\/\/link.ovr.ai\/map2earn\"><span style=\"font-weight: 400;\">https:\/\/link.ovr.ai\/map2earn<\/span><\/a><b><\/b><\/p>\n<p><b>2. Upload<\/b><span style=\"font-weight: 400;\"> your footage and let the OVER pipeline reconstruct the scene, Gaussian splat, collision mesh, metric scale, packaged as one USDZ<\/span><\/p>\n<p><b>3. Download<\/b><span style=\"font-weight: 400;\"> your USDZ file<\/span><\/p>\n<p><b>4. Import<\/b><span style=\"font-weight: 400;\"> into NVIDIA Isaac Sim or Isaac Lab, and start training<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">It&#8217;s completely free. And because captures on OVER feed the map2earn\u2122 program, the same walk that builds your simulation environment can also earn you rewards.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">The same export capability is also planned for <\/span><a href=\"https:\/\/www.freegaussian.ai\/\"><span style=\"font-weight: 400;\">https:\/\/www.freegaussian.ai\/<\/span><\/a><span style=\"font-weight: 400;\">, bringing simulation-ready output to browser-based reconstructions.\u00a0<\/span><\/p>\n<p>&nbsp;<\/p>\n<h2><span style=\"font-weight: 400;\">From Map2Earn\u2122 to Physical AI Infrastructure<\/span><\/h2>\n<p><span style=\"font-weight: 400;\">For our community, this launch means something bigger.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">For years, the OVER community has been mapping the world through map2earn\u2122, producing what is now more than <\/span><b>270,000 mappings<\/b><span style=\"font-weight: 400;\"> and over <\/span><b>105 million images<\/b><span style=\"font-weight: 400;\"> of real-world locations, a dataset already licensed by one of the world&#8217;s most valuable companies to train next-generation vision models.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">USDZ export makes the value of that activity tangible in a whole new industry. Every scene captured by a mapper isn&#8217;t just AR real estate or VPS coverage anymore. It&#8217;s potential training infrastructure for Physical AI, the environments where tomorrow&#8217;s robots will learn to see, navigate, and act.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">This is the DePIN flywheel we described in our 2026 roadmap, turning: robotics companies need real-world environments at scale; our community can capture them with devices they already own; and the demand flows back into the ecosystem that produced the data. Mappers earn. Robots learn.<\/span><\/p>\n<p>&nbsp;<\/p>\n<h2><span style=\"font-weight: 400;\">The World Is the Best Training Ground<\/span><\/h2>\n<p><span style=\"font-weight: 400;\">Our mission has always been to make the physical world machine-readable, first for people, through AR and navigation, and now for machines, through robotics simulation.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">The industry is converging on USDZ and OpenUSD as the interchange format for simulation, and NVIDIA Isaac Sim has become the reference platform for training embodied AI. With today&#8217;s launch, the bridge between a real place and a robot&#8217;s training environment is a five-minute capture and a free download.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">You can try it now in the OVER app <\/span><a href=\"https:\/\/link.ovr.ai\/map2earn\"><span style=\"font-weight: 400;\">https:\/\/link.ovr.ai\/map2earn<\/span><\/a><span style=\"font-weight: 400;\"> or on the web <\/span><a href=\"https:\/\/marketplace.ovr.ai\/mappings\"><span style=\"font-weight: 400;\">https:\/\/marketplace.ovr.ai\/mappings<\/span><\/a><span style=\"font-weight: 400;\">\u00a0<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Join the conversation on <\/span><a href=\"https:\/\/discord.gg\/overthereality\"><span style=\"font-weight: 400;\">Discord<\/span><\/a><span style=\"font-weight: 400;\">, and if you&#8217;re a robotics team looking for environment datasets at scale, reach out at business@ovr.ai.<\/span><\/p>\n","permalink":"usdz","date":"2026-07-17 15:17:59","image_small":"https:\/\/blog.ovr.ai\/wp-content\/uploads\/2026\/07\/Screenshot-2026-07-17-at-17.07.46_11zon-150x150.jpg","image_medium":"https:\/\/blog.ovr.ai\/wp-content\/uploads\/2026\/07\/Screenshot-2026-07-17-at-17.07.46_11zon-300x168.jpg","image_large":"https:\/\/blog.ovr.ai\/wp-content\/uploads\/2026\/07\/Screenshot-2026-07-17-at-17.07.46_11zon-1024x574.jpg","image_full":"https:\/\/blog.ovr.ai\/wp-content\/uploads\/2026\/07\/Screenshot-2026-07-17-at-17.07.46_11zon-scaled.jpg","single_url":"https:\/\/blog.ovr.ai\/usdz\/","translations":{"en":{"single_url":"https:\/\/blog.ovr.ai\/usdz\/","permalink":"usdz"},"fr":{"single_url":"https:\/\/blog.ovr.ai\/usdz\/","permalink":"usdz"},"es":{"single_url":"https:\/\/blog.ovr.ai\/usdz\/","permalink":"usdz"},"tr":{"single_url":"https:\/\/blog.ovr.ai\/usdz\/","permalink":"usdz"},"zh":{"single_url":"https:\/\/blog.ovr.ai\/usdz\/","permalink":"usdz"}}}]}