ALIENWARS GYM
PROCEDURAL WORLDS / REINFORCEMENT LEARNING

A gym for training AlienWars agents.

Build a world. Command a fleet. See what the agents sense. Twelve vehicles share a 3D environment, with A* global routes and five trained PPO policies for local navigation.

Explore Map Lab ↗Try a guided experimentSee training results
PufferLib 5Native C/CUDA reinforcement learning
Raylib3D terrain, vehicles and sensor rendering
WebAssembly RuntimeSimulation and policy inference in your browser

59-second showcase · Silent · Recorded September 15, 2026. Explore the current build below.

LIVE IN THE BROWSER

One seed. A world of decisions.

Wave Function Collapse connects terrain, road grades and tunnel profiles. Irregular oval coastlines, graded beaches, lakes and hills shape the battlefield. Choose a coherent Temperate, Desert or Frozen palette, then explore symmetric and asymmetric layouts.

Grown bodies. Living materials.

All twelve fleet roles now have distinct biological bodies, with grown nursery forms at both bases. Temperate fungal parasols, Desert mushroom-cacti and Frozen frost harps replace traditional trees. Four 1K material scans, deeper shadows, moving shore foam and boat wakes add physical detail. Play view gives the battlefield more room.

Biological scout and alien fungi, with textured Temperate terrain

Twelve living bodies share five movement families. Cultivation and combat remain future work.

Explore in Play view ↗

Look inside a living world.

Open the official Flecs Explorer beside the battlefield. Inspect live body state, mission progress, sensor equipment and PPO observations from the same running simulation.

Official Flecs Explorer displaying live unit components beside the Temperate battlefield
Inspect the live Flecs world ↗

Take the controls.

  1. Move around: drag to pan, Shift-drag to orbit, scroll to zoom.
  2. Command: click a vehicle, or select its family, then right-click a destination.
  3. Observe: follow a unit, toggle its sensors, or reveal the underground network.
  4. Regenerate: choose a palette and create a new symmetric or asymmetric world.
Open the live gym ↗
SHARED-WORLD PPO

Five policies learn together.

Ground vehicles, boats, quadcopters, fixed wings and submarines train in the same worlds. Each vehicle keeps its own goal and recurrent memory.

49.5 million steps

Three training seeds, recorded metrics and held-out evaluation. Local navigation remains experimental; combat and tactical coordination are future work.