Evolve Control Console

Episode 1

Best Fitness 0.0

Accuracy 0%

Rolling Accuracy 0%

Successes 0

Mistakes 0

Active Agents 10/10

Ticket Source SYNTHETIC

Classifier Acc —

Misled by AI —

How to Use This Simulation

Guide the agents as they learn to sort tickets into the correct goal zones. Watch the rolling accuracy and best fitness to see learning progress.

Use Simulation Pace and Episode Duration to speed up training.
Adjust Population Size and Ticket Count to change difficulty.
Drag agents, the nest, or goal zones to reposition the environment.

Performance History (Best Agent)

Active Policy Network (Best Agent)

Live Neural Network Activity (Red=Negative, Green=Positive)

Simulation Controls

Simulation Pace

Game Speed: 3x

Episode Duration (Frames): 6000

Manual Episode (Ends on Restart)

Agents & Tickets

Population Size: 10

Ticket Count: 50

Ticket Goal Buffer: 120px

Ticket Spawn Layout

Ticket Category Grouping

Ticket Source

Where categories come from

Goal Zones

Goal Size: 100%

Incident Goal Edge

Service Req Goal Edge

Problem Goal Edge

Learning (NEAT-lite)

Steps / Gen: 6000

Hidden Nodes: 20

Mutation %: 22%

Elite ratio: 20%

Mutate weights: 0.25

Mutate biases: 0.25

Crossover bias: 50% A

Selection pool: ×2.0

Fresh spawn %: 6%

Auto next gen on

Penalties

Wall Penalty: 0

Stalled Penalty: 0

Wrong Zone Penalty: -1

Reward Shaping

Movement Reward: 0.00

Curiosity Reward: 0.00

Progress Scale: 0.00

Progress Streak Reward: 0.00

Pickup Reward: 0

Delivery Base Reward: 1

Delivery Energy Bonus: 0

Near Correct Goal Reward: 0.00

Near Wrong Goal Penalty: -0.00

Target Switch Penalty: -0.00

Game Physics / ACO

Sensor Range: 180px

Battery Drain: Normal

Pheromone Decay: 0.990

Visuals

Show Policy Network Show Sensors (Simplified) Show Pheromone Heatmap Show Particles (Feedback) Show Best Agent Trail

Evolve: Data Sorting Simulation JEV OFF — random categories FPS: 0

EPISODE 1

[SPACE] Pause | [S] Restart Episode | Click & Drag Agent / NEST / Zone