notebook.community

Edit and run



In [7]:

    
%load_ext autoreload
%autoreload 2
%matplotlib inline









    



The autoreload extension is already loaded. To reload it, use:
  %reload_ext autoreload



In [8]:

    
import numpy as np
import tempfile
import tensorflow as tf

from tf_rl.controller import DiscreteDeepQ, HumanController
from tf_rl.controller import GreedyController, RandomController, WeightedRandomController
from tf_rl.simulation import KarpathyGame
from tf_rl import simulate
from tf_rl.models import MLP

from __future__ import print_function



In [22]:

    
current_settings = {
    'objects': [
        'friend',
        'enemy',
    ],
    'colors': {
        'hero':   'yellow',
        'friend': 'green',
        'enemy':  'red',
    },
    'object_reward': {
        'friend': 0.1,
        'enemy': -0.1,
    },
    'hero_bounces_off_walls': False,
    'world_size': (700,500),
    'hero_initial_position': [400, 300],
    'hero_initial_speed':    [0,   0],
    "maximum_speed":         [50, 50],
    "object_radius": 10.0,
    "num_objects": {
        "friend" : 25,
        "enemy" :  25,
    },
    "num_observation_lines" : 32,
    #"observation_line_length": 120.,
    "observation_line_length": 80.,
    "tolerable_distance_to_wall": 50,
    "wall_distance_penalty":  -0.0,
    "delta_v": 50
}



In [23]:

    
# Greedy human controller
# create the game simulator
g = KarpathyGame(current_settings)
current_controller = GreedyController({b"w": 3, b"d": 0, b"s": 1,b"a": 2,}, current_settings["num_observation_lines"])



In [24]:

    
FPS          = 30
ACTION_EVERY = 3
    
fast_mode = True
if fast_mode:
    WAIT, VISUALIZE_EVERY = False, 1
else:
    WAIT, VISUALIZE_EVERY = True, 1

    
try:
    with tf.device("/cpu:0"):
        simulate(simulation=g,
                 controller=current_controller,
                 fps=FPS,
                 visualize_every=VISUALIZE_EVERY,
                 action_every=ACTION_EVERY,
                 wait=WAIT,
                 disable_training=True,
                 simulation_resolution=0.001,
                 max_frame=3000,
                 save_path=None)
except KeyboardInterrupt:
    print("Interrupted")









    














    



('Reach max frames ', 3000)

Visualizing what the agent is seeing

Starting with the ray pointing all the way right, we have one row per ray in clockwise order. The numbers for each ray are the following:

first three numbers are normalized distances to the closest visible (intersecting with the ray) object. If no object is visible then all of them are $1$. If there's many objects in sight, then only the closest one is visible. The numbers represent distance to friend, enemy and wall in order.
the last two numbers represent the speed of moving object (x and y components). Speed of wall is ... zero.

Finally the last two numbers in the representation correspond to speed of the hero.



In [31]:

    
g.__class__ = KarpathyGame
np.set_printoptions(formatter={'float': (lambda x: '%.2f' % (x,))})
x = g.observe()
new_shape = (x[:-2].shape[0]//g.eye_observation_size, g.eye_observation_size)
print(x[:-4].reshape(new_shape))
print(x[-4:])
g.to_html()









    



[[1.00 1.00 1.00 0.00 0.00]
 [1.00 1.00 1.00 0.00 0.00]
 [1.00 1.00 1.00 0.00 0.00]
 [1.00 1.00 1.00 0.00 0.00]
 [1.00 1.00 1.00 0.00 0.00]
 [1.00 1.00 1.00 0.00 0.00]
 [1.00 1.00 1.00 0.00 0.00]
 [1.00 1.00 1.00 0.00 0.00]
 [0.55 1.00 1.00 -0.55 0.26]
 [1.00 1.00 1.00 0.00 0.00]
 [0.61 1.00 1.00 0.46 -0.18]
 [1.00 0.37 1.00 -0.71 -0.08]
 [1.00 0.41 1.00 -0.71 -0.08]
 [1.00 1.00 1.00 0.00 0.00]
 [1.00 1.00 1.00 0.00 0.00]
 [1.00 0.88 1.00 -0.80 -0.28]
 [1.00 1.00 1.00 0.00 0.00]
 [1.00 1.00 1.00 0.00 0.00]
 [1.00 1.00 1.00 0.00 0.00]
 [1.00 1.00 1.00 0.00 0.00]
 [1.00 0.72 1.00 -0.90 -0.56]
 [1.00 1.00 1.00 0.00 0.00]
 [1.00 1.00 1.00 0.00 0.00]
 [1.00 1.00 1.00 0.00 0.00]
 [1.00 1.00 1.00 0.00 0.00]
 [1.00 1.00 1.00 0.00 0.00]
 [1.00 1.00 1.00 0.00 0.00]
 [1.00 1.00 1.00 0.00 0.00]
 [1.00 1.00 1.00 0.00 0.00]
 [1.00 1.00 1.00 0.00 0.00]
 [1.00 0.64 1.00 -0.64 0.43]
 [1.00 0.63 1.00 -0.64 0.43]]
[0.00 0.00 0.14 0.20]






    Out[31]:



In [ ]: