Gizmo generates editable 3D environments for robot training from text and images (runtimewire.com)

Antim Labs has launched the public beta of Gizmo, an innovative AI tool that transforms text and reference images into editable 3D environments tailored for robot training. This product positions itself as a "world authoring layer for simulations," allowing robotics teams to quickly create and iterate on training environments without the painstaking manual 3D modeling processes typically required. Gizmo generates structured assets based on geometric primitives, enabling easier modifications, such as resizing or changing object properties, rather than relying on fixed polygon meshes. What makes Gizmo particularly significant for the AI/ML community is its ability to streamline the simulation process across multiple robotics platforms. By exporting environments in formats compatible with various simulation engines—such as USD for Nvidia's Isaac Sim, MJCF for Google DeepMind's MuJoCo, and SDF for Gazebo—teams can efficiently test and compare performance across different platforms without creating separate versions of each environment. The API further facilitates automation in robotics development, allowing coded agents to request and integrate new simulation assets on-the-fly. With the potential to improve the realism and efficiency of robot training through precise physics and streamlined workflows, Gizmo aims to be a game changer in the realm of robotics simulation.