Planetary

simulation environments for spatial AI

A procedural world generator with semantic object system, environment randomization, and an RL training interface.

Preview to World in one click

Use cases

  • Train physical AI, such as humanoid robots and drones
  • Test out agentic avatars in a game-like environment

Features

  • Create procedurally generated infinite worlds from noise functions
  • Instantly add procedural assets with semantic labelling and affordances
  • Train RL agents in headless simulations sampled from a world distribution, for thousands of variations of the same 3D space

Upcoming

  • Full in-house procedural asset library, all equipped with affordances
  • Headless RL mode for parallel training
  • Accurate weather and environmental conditions
  • Buildings and cities

Releasing 2026