PixelCopter REINFORCE Agent — V3
Project
Training a REINFORCE policy-gradient agent to play PixelCopter.
Environment
- Environment: Pixelcopter-PLE-v0
- Algorithm: REINFORCE
- Network: 7 → 64 → 128 → 2
- Optimizer: Adam
- Learning rate: 0.0001
- Gamma: 0.99
Training
- Training episodes: 5000
Evaluation
- Evaluation episodes: 30
- Mean reward: 11.93
- Standard deviation: 11.51
- Mean minus standard deviation: 0.42
Evaluation status
Course target: Mean reward - Standard deviation >= 5.
Current score: 0.42
Target not achieved yet. Further training and evaluation are required.