Foodsim Arena

Multi-agent NEAT sandbox where four collector teams and one blocker team gather fruit, dodge obstacles, and evolve tactics.

Open Foodsim

BallLab

Neuroevolution meets physics: watch agents learn to keep bouncing and survive.

Open BallLab

Snake AI

Grid-based pathfinding: watch the snake route with BFS, Hamilton cycles, and endgame safeguards.

Open Snake AI

VolleySim (NEAT)

Cooperative NEAT volleyball where two agents learn to keep the ball alive.

Open VolleySim

Data Sorting Lab

Hybrid Policy Learning experiment that visualizes sorting and routing behavior.

Open Data Sorting

How NEAT Works

Plain-English guide that explains the FoodSim and NEAT mechanics.

Open NEAT guide

Deep dives into the core experiments

These are the most popular simulations. Each summary explains the algorithm, how to work with the controls, and what to explore while the agents learn.

Snake AI

Snake AI uses deterministic pathfinding on a grid. The snake tracks food, its own body, and open space to choose safe routes and survive late-game.

  • Algorithm: BFS for shortest paths, A+-style heuristic scoring, Hamilton cycles for full-board coverage, and endgame tail-follow logic.
  • Work with it: toggle overlays to see the BFS/Hamilton routes and switch playback speed to study decisions.
  • Explore: watch how the planner pivots to Hamilton routes as the board fills and endgame safety kicks in.

Foodsim Arena

Foodsim is a multi-agent NEAT arena. Teams evolve neural networks that compete for food, avoid hazards, and develop emergent tactics.

  • Algorithm: NEAT (NeuroEvolution of Augmenting Topologies) with population fitness scoring.
  • Work with it: run up to four collector teams plus one blocker team, toggle obstacles, and compare team fitness charts.
  • Explore: blockers harass rivals and knock carried fruit loose, while collectors specialize into scouts and harvesters.
  • Music: pick a track from the music panel and adjust the volume to set the mood.

Data Sorting Lab

The Data Sorting demo visualizes agents routing items to the correct lanes, showing how learned policies organize data streams.

  • Algorithm: hybrid policy learning that blends heuristic routing with learned action selection.
  • Work with it: feed more items, slow the simulation, and watch the routing decisions per lane.
  • Explore: inspect how the policy adapts as the input distribution changes.

BallLab

BallLab evolves bouncing agents that must keep moving without crashing. It is a compact physics sandbox for neuroevolution.

  • Algorithm: NEAT-based evolution with survival time and stability as fitness signals.
  • Work with it: reset the population, then tweak gravity or damping if controls are available.
  • Explore: observe how agents learn to conserve momentum and avoid chaotic collisions.

Vollyball Showdown

Vollyball pits cooperative agents against physics and timing challenges. The goal is to keep the ball in play and coordinate movement.

  • Algorithm: NEAT-style cooperative evolution with shared rewards for successful volleys.
  • Work with it: run multiple matches, then compare how coordination improves over generations.
  • Explore: look for synchronized jumps and positioning as the agents learn spacing.

Tips for the best experience

Most simulations respond instantly, but giving them a few moments to run reveals richer behavior. If a demo has controls, try changing one value at a time to see clear differences.