AI Learns to Drive - Changelog 0.10.2 (New Outputs & Rewards)

Steam post imageSteam post image

NEW OUTPUTS

Added optional neural network outputs:

• Handbrake: Apply handbrake for drifting Train your AI to drift!

• Brakes: Apply regular brakes

REWARD SYSTEM OVERHAUL

• Reward Editor: Customize rewards and choose what behaviors to train

• 7 NEW Reward Metrics:

• Drifting: Time spent in a sustained drift

• Actual Distance: Actual distance the car's body has traveled

• Wall Distance: Distance from the nearest wall

• Center Distance: Distance from the track's center line

• Momentum Change: How much the car's speed changes

• Throttle Frequency: How much the throttle input changes

• Braking Frequency: Total time spent braking

1. 2 Reward Types:

• Default: Total accumulated value over the entire run

• Average: Average value per second while active

EVOLUTION EDITOR IMPROVEMENTS

• Output Management: Add or remove brake and handbrake outputs from trained networks

• AddedSmart Weight EditingExisting weights preserved when editing trained networks New weights initialized to 0
• AdjustedSensor EditingImproved behavior when adjusting sensor degrees on trained networks

• Fixed: Snapshot system

MANUAL DRIVING

• Fixed: Focus issues during driving

• Brake Input (Space) + Handbrake Input: (B Key)

OTHER IMPROVEMENTS

• Precision Controls: More decimal places for weight decay and mutation rate settings

• Network Preview: Display more information when selecting networks

• AdjustedReward NormalizationChanged from normal distribution to uniform distribution