charvest/wrap.py at main · rtyasdf/charvest · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
from ctypes import *
import numpy as np
import os
from time import perf_counter as pc
from collections import namedtuple


dir_path = os.path.dirname(__file__)
lib = CDLL(dir_path + "/env.so")

char_p = POINTER(c_char)
char_pp = POINTER(char_p)
int_p = POINTER(c_int)
int_pp = POINTER(int_p)
float_p = POINTER(c_float)

class Pair(Structure):
  _fields_ = [("x", c_int),
              ("y", c_int)]

pair_p = POINTER(Pair)


class HarvestAgent(Structure):
  _fields_ = [("id", c_char),
              ("pos", Pair),
              ("orientation", c_char),
              ("reward", c_int)]


class AppleMap(Structure):
  _fields_ = [("pos", pair_p),
              ("size", c_int),
              ("apple_symbol", c_char),
              ("id", int_pp),
              ("neighbors", int_pp),
              ("alive", int_p),
              ("dead", int_p),
              ("new_head", int_p),
              ("old_head", int_p),
              ("hide", int_p)]


class HarvestEnv(Structure):
  _fields_ = [("num_of_agents", c_int),
              ("world_map", char_pp),
              ("apple_map", POINTER(AppleMap)),
              ("spawn_pos", pair_p),
              ("wall_pos", pair_p),
              ("spawn_pos_size", c_int),
              ("wall_pos_size", c_int),
              ("agents", POINTER(HarvestAgent)),
              ("dead_agents", char_p)]


Seed = lib.seed
Seed.argtypes = [c_int]


Create = lib.create_env
Create.restype = HarvestEnv
Create.argtypes = [c_int]


Reset = lib.reset
Reset.argtypes = [HarvestEnv, float_p, float_p, float_p, float_p, float_p]


Step = lib.step
Step.argtypes = [HarvestEnv, int_p, float_p, float_p, float_p, float_p, float_p, float_p]

DIAMETER = c_int.in_dll(lib, 'DIAMETER').value
NUM_OF_ROWS = c_int.in_dll(lib, 'NUM_OF_ROWS').value
MAP_ROW_LENGTH = c_int.in_dll(lib, 'MAP_ROW_LENGTH').value

Observation = namedtuple('Observation', 'full_map local_obs position orientation able_to_shoot')


class Env:
  def __init__(self, n_agents: int, n_steps: int=1000, seed: int=0):
    if seed:
      Seed(seed)

    self.c_env = Create(n_agents)
    self.n_steps = n_steps
    self.n_agents = n_agents
    self.create_new_arrays()


  def create_new_arrays(self):
    self.full_map = np.zeros((self.n_steps + 1, NUM_OF_ROWS, MAP_ROW_LENGTH), dtype=np.float32)
    self.obs = np.zeros((self.n_steps + 1, self.n_agents, DIAMETER, DIAMETER), dtype=np.float32)

    self.positions = np.zeros((self.n_steps + 1, self.n_agents, 2), dtype=np.float32)
    self.orientations = np.zeros((self.n_steps + 1, self.n_agents), dtype=np.float32)
    self.able_to_shoot = np.zeros((self.n_steps + 1, self.n_agents), dtype=np.float32)

    self.reward = np.zeros((self.n_steps + 1, self.n_agents), dtype=np.float32)


  def reset(self):
    self.t = 0
    self.create_new_arrays()

    Reset(self.c_env,
          self.full_map[self.t].ravel().ctypes.data_as(float_p),
          self.obs[self.t].ravel().ctypes.data_as(float_p),
          self.positions[self.t].ravel().ctypes.data_as(float_p),
          self.orientations[self.t].ctypes.data_as(float_p),
          self.able_to_shoot[self.t].ctypes.data_as(float_p))

    return Observation(self.full_map[self.t], self.obs[self.t],
                       self.positions[self.t], self.orientations[self.t], self.able_to_shoot[self.t])


  def step(self, actions):
    self.t += 1
    Step(self.c_env, actions.ctypes.data_as(int_p),
         self.full_map[self.t].ravel().ctypes.data_as(float_p),
         self.obs[self.t].ravel().ctypes.data_as(float_p),
         self.positions[self.t].ravel().ctypes.data_as(float_p),
         self.orientations[self.t].ctypes.data_as(float_p),
         self.able_to_shoot[self.t].ctypes.data_as(float_p),
         self.reward[self.t].ctypes.data_as(float_p))

    obs = Observation(self.full_map[self.t], self.obs[self.t],
                      self.positions[self.t], self.orientations[self.t], self.able_to_shoot[self.t])

    return obs, self.reward[self.t], self.t == self.n_steps


if __name__ == "__main__":
  n_agents = 5
  n_episodes = 100

  e = Env(n_agents)

  actions = np.zeros((n_agents, ), dtype=np.int32)

  T = pc()
  for i in range(n_episodes):
    obs = e.reset()
    done = False
    while not done:
      obs, rew, done = e.step(actions)

  T = pc() - T

  print("Full map:")
  print(obs.full_map)

  for n in range(n_agents):
    print(f"\tAgent {n}:")
    print(obs.local_obs[n])
    print("\n\n")

  print(f"{n_episodes} episodes takes {T : .2f} seconds")