Splatica

Splatica is a United Kingdom-based physical AI startup whose platform converts 360-degree video — notably footage captured with Insta360 cameras and Antigravity drones — into photorealistic, physically accurate 3D digital twins. The company positions itself at the intersection of computer vision, 3D reconstruction and robotics simulation, offering a rapid, largely automated pipeline that transforms raw .INSV footage into interactive 3D scenes that can be explored, exported and shared. Splatica's core commercial proposition is data generation for physical AI. The company argues that the biggest challenge for physical AI is diverse data collection, and it addresses this by providing a fast pipeline for creating photorealistic, physically accurate environments. Its digital 3D replicas are designed for physical AI training, and the platform transfers physically accurate 3D splat scenes into NVIDIA Isaac Sim and Omniverse, ready for native prompts and model-training pipelines. The workflow is presented in four stages: object segmentation (segment anything, extract and replace objects with known USD assets); collision meshes and USD physics (extracted with a vision-language model and baked into the USD scene together with separate objects and a scene graph); minimizing the sim-to-real gap (training robots in a reliable digital copy of a physical space to reduce training in the real environment); and photorealism (realistic representation where camera intrinsics can be simulated in any required format, with video data collection for training VLA and SLAM models). For end users, Splatica offers a simple three-step process: capture data by recording video with an Insta360 camera or Antigravity drone; upload and process the raw footage (just an .INSV file) for automated 3D reconstruction; then view, explore and export the resulting 3D model. The company showcases community-created scenes and an interactive 3D scan of an industrial factory in Turkey, with a viewer that can display the volumetric point cloud. Splatica is privately held and operates globally through its web platform. Its public materials emphasize partnerships, community-generated scenes and creator videos, and it markets itself as a rapid, hassle-free route from 360 video to stunning 3D models.

Splatica is a physical AI startup that emerged around the commercial need to generate diverse, photorealistic training data for robots and embodied AI systems. Its public identity is built on a single core capability: converting 360-degree video into simulation-ready 3D digital twins. The company describes itself as a rapid pipeline for creating photorealistic, physically accurate environments, addressing what it calls the biggest challenge for physical AI — diverse data collection. The platform was designed around consumer and prosumer 360 capture hardware, particularly Insta360 cameras and Antigravity drones. Users record video, upload the raw .INSV file, and receive an automatically reconstructed 3D model that can be explored, exported and shared. This lowers the barrier to entry for creating 3D scenes, moving the process away from specialist photogrammetry workflows toward a largely automated, web-based service. As the company developed, it extended its output beyond visual 3D models into simulation-ready assets. Splatica transfers physically accurate 3D splat scenes into NVIDIA Isaac Sim and Omniverse, ready to use with native prompts and model-training pipelines. Its pipeline includes object segmentation, collision mesh and USD physics extraction using a vision-language model, and baking into a USD scene with separate objects and a scene graph. This makes the output usable for robot training, with the explicit goal of minimizing the sim-to-real gap by letting robots train in a reliable digital copy of a physical space. The company also emphasizes photorealism and camera simulation, noting that camera intrinsics can be simulated in any required format, and that its scenes support video data collection for training VLA and SLAM models. It showcases community-created scenes and an interactive 3D scan of an industrial factory in Turkey, with a viewer that can display the volumetric point cloud. Splatica is privately held and operates globally through its website. Its public materials highlight partnerships, community scenes and creator videos, and it markets itself as a fast, hassle-free route from 360 video to stunning 3D models. The company's Wikidata entry describes it as a physical AI startup converting 360 video into simulation-ready 3D digital twins.

A rapid, automated pipeline for creating photorealistic, physically accurate 3D environments for physical AI and robotics simulation.

• Splatica 3D reconstruction platform

  Web-based pipeline that turns Insta360 .INSV footage into interactive 3D digital twins.

• Simulation-ready digital replicas

  Physically accurate 3D splat scenes transferred into Isaac Sim and Omniverse for physical AI training.

• Object segmentation and USD physics

  Segment anything, extract and replace objects with known USD assets; collision meshes and USD physics extracted with VLM and baked into the USD scene.

1. 1Splatica: Immersive 3D Reconstructionssplatica.com
2. 2Splatica - Wikidatawikidata.org