How It Works
- Watch the agent act
- Press / for "good" or z for "bad"
- The agent learns your preferences
That's it. The agent builds a model of what you like and adjusts its behavior accordingly.
TAMER works best when you understand the task. Before training, try controlling the agent yourself for a few episodes using Human Control (press m to toggle, then use arrow keys or i/j/k/l). This helps you know what good behavior looks like.
Keys
Feedback
/ good
z bad
Controls
Space toggle training
1 start/pause
m toggle human ctrl