RL for ChungLi Traffic 400 tiers

 市區交通模擬，如何控制燈號以增進運輸量 (RL)

    Birds' eye view and follow mode

 (credit: 與專題生共同創作)

sample, share

🚦 RL Traffic Control — Train, Save, Compare

Agents persist across sessions • Continue training anytime

🎮 Training (1747 episodes)

1x5x10x20x

1747

Episodes

5%

Exploration

1358079

Q-Table

-303.5

Avg R (10)

Learning Curve

🚦 RL Traffic Control — Train, Save, Compare

Agents persist across sessions • Continue training anytime

📊 Comparison Complete

Metric

❌ Untrained

✅ Trained (1688 eps)

Total Reward

-359.8

-158.8(+56%)

Throughput

17

21

Avg Wait

44.3

29.2

✓ Learning Verified! Trained agent outperformed random.