1
Launch the demo as administrator.
2
To train your bitbot you have to complete following steps:
•
•
Record a demo play for your BitBot. Recommended length – 10-15 minutes of gameplay.
•
•
Once finished – proceed to the training section. Here you can tweak training options for your BitBot. There are two main sections:
•
•
o
Presets – train your BitBot using predefined settings.
•
•
o
Advanced options - advanced control over training parameters. Advanced options include sections:
•
•
•
.
Rewards – Defines the strength of the rewards (be it a negative or positive) for a specific behavior. Based on these settings BitBot will try to achieve the maximum positive reward. Be aware that setting rewards too high or too low can confuse BitBot resulting in a bad training session.
•
•
•
.
Hyperparameters – The most crucial part of the training where you specify Machine Learning parameters. Every parameter and its combinations can lead to a significant change in training stability and results. You can learn more about each parameter by hovering over it or by following the link https://github.com/Unity-Technologies/ml-agents/blob/main/docs/Training-Configuration-File.md
•
•
•
.
Choose a recorded demo from the list. By default it will be the last recorded demo.
•
•
•
.
Select a number of simultaneous training sessions. The specs of your PC define this parameter. If you have a high-end PC with 32GB RAM or more and the latest GPU, we recommend 6 or more training sessions. If you run into freezes of PC or other errors try to lower this number.
•
•
•
.
Start training by pressing the start button.
•
•
Training process can take from 20 minutes to a couple hours depending on the hardware it’s running and settings chosen for training. It's your decision when to stop the training process. To stop training click “Stop training” button.
•
•
You can monitor training progress by clicking the “Show progress” button. This will open up tensorflow graphs. The most important parameter to watch is the Mean Reward. It must grow to the maximum positive value. If it doesn’t grow or is unstable try changing rewards, training configuration, or demo.
3
After training is complete you can start competing with other BitBots ( for this demo it is our pre-trained BitBot). Enter the Bot vs Bot section, choose your BitBot from the list and start the match! The first BitBot to complete 3 laps is a winner.