Reinforcement Learning Logo Turtle
This package wraps Python's built-in Logo Turtle.
The problem with the built-in turtle is that it's not easy to get the drawn canvas as a numeric array. It's possible after lots of effort, but then you'd find that it was very slow.
This package wraps/overrides some of the built-in turtle methods to keep a (*approximate) copy of the drawn canvas as a NumPy PyTorch array. I say approximate because the arrow I use as the turtle itself (a 10x10 icon) roughly looks like the one used in the native package, but isn't the same. It doesn't matter.
Screenshots
What's this for?
Python's turtle makes a great testing ground for reinforcement learning / deep-Q learning algorithms, or it would if we could easily get at the array/data/rendered image underlying the window in which the turtle shows. So, this makes that array available.
Usage
from rlturtle import RLTurtle
tortoise = RLTurtle(200,200, n_history_frames=2)
tortoise.forward(10)
tortoise.plot()
tortoise.left(90)
tortoise.forward(10)
current_frame = tortoise.canvas
prev_frame = tortoise.prev_frames[1]
older_frame = tortoise.prev_frames[0]
...
model.act(torch.cat((current_frame, prev_frame, older_frame)))