GGDDPG/wrapper.py at main · m-coder-stack/GGDDPG · 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
import gymnasium as gym
from gymnasium import ObservationWrapper, ActionWrapper
from gymnasium import spaces
import numpy as np
import copy

# Function to calculate the dimensions of each subspace
def calculate_space_dimensions(space):
    if isinstance(space, gym.spaces.Dict):
        # Sum dimensions of all subspaces for Dict space
        return sum(calculate_space_dimensions(subspace) for subspace in space.spaces.values())
    elif isinstance(space, gym.spaces.Discrete):
        # Discrete space has a single dimension
        return 1
    elif isinstance(space, gym.spaces.Box):
        # Product of the Box shape's dimensions
        return np.prod(space.shape)
    elif isinstance(space, gym.spaces.MultiBinary):
        # MultiBinary space dimension is given by 'n'
        return space.n
    elif isinstance(space, gym.spaces.Tuple):
        # Sum dimensions of all subspaces for Tuple space
        return sum(calculate_space_dimensions(subspace) for subspace in space.spaces)
    # Additional space types can be added here as needed
    else:
        raise TypeError("Unsupported space type")


class RelativePosition(ObservationWrapper):
    def __init__(self, env):
        super().__init__(env)
        # modify the observation space
        # if the observation space is a dictionary
        # remove the center position from the observation space
        if isinstance(env.observation_space, spaces.Dict):
            obs_relay_position = env.observation_space.spaces["relay"]["position"]
            obs_relay_height = env.observation_space.spaces["relay"]["height"]
            obs_client = env.observation_space.spaces["client"]
            obs_center = env.observation_space.spaces["center"]
            self.observation_space = spaces.Dict({
                "relay": spaces.Dict({
                    "position": spaces.Box(low=obs_relay_position.low.min() - obs_center.high.max(), \
                                           high=obs_relay_position.high.max() - obs_center.low.min(), \
                                              shape=obs_relay_position.shape, dtype=np.float32),
                    "height": obs_relay_height,
                }),
                "client": spaces.Box(low=obs_client.low.min() - obs_center.high.max(), \
                                     high=obs_client.high.max() - obs_center.low.min(), \
                                        shape=obs_client.shape, dtype=np.float32),
            })

        else:
            self.observation_space = env.observation_space

    def observation(self, observation):
        return {
            "relay": {
                "position": observation["relay"]["position"] - observation["center"],
                "height": observation["relay"]["height"],
            },
            "client": observation["client"] - observation["center"],
        }

class FlattenDict(ObservationWrapper):
    def __init__(self, env):
        super().__init__(env)
        assert isinstance(env.observation_space, spaces.Dict), "environment's observation space must be a Dict"

        # calculate the size of new observation space
        obs_dim = calculate_space_dimensions(env.observation_space)
        obs_client: spaces.Box = env.observation_space.spaces["client"]
        obs_relay_position: spaces.Box = env.observation_space.spaces["relay"]["position"]
        obs_relay_height: spaces.Box = env.observation_space.spaces["relay"]["height"]

        space_low = np.concatenate([obs_relay_position.low, obs_relay_height.low], axis=1)
        space_low = np.concatenate([space_low, obs_client.low], axis=None)

        space_high = np.concatenate([obs_relay_position.high, obs_relay_height.high], axis=1)
        space_high = np.concatenate([space_high, obs_client.high], axis=None)

        # create new observation space
        self.observation_space = spaces.Box(low=space_low, high=space_high, shape=(obs_dim,), dtype=np.float32)

    def observation(self, observation):
        # tranform the observation from dictionary to one-dimensional array
        obs_relay_position = observation["relay"]["position"]
        obs_relay_height = observation["relay"]["height"]
        obs_client = observation["client"]

        obs_relay = np.concatenate([obs_relay_position, obs_relay_height], axis=1)
        return np.concatenate([obs_relay, obs_client], axis=None)
        # return np.concatenate([obs.flatten() for obs in observation.values()])


class SerializeAction(ActionWrapper):
    def __init__(self, env: gym.Env, is_polar: bool = True):
        super().__init__(env)
        # assert isinstance(env.action_space, spaces.Dict), "environment's action space must be a Dict"

        self.is_polar = is_polar

        if is_polar == False:
            # calculate the size of new action space
            act_dim = calculate_space_dimensions(env.action_space)
            # create new action space
            self.action_space = spaces.Box(low=env.action_space.low.flatten(), high=env.action_space.high.flatten(), shape=(act_dim,), dtype=np.float32)
        else:
            # calculate the size of new action space
            act_dim = calculate_space_dimensions(env.action_space["displacement"]) + calculate_space_dimensions(env.action_space["direction"])
            # create new action space
            space_low = np.concatenate([env.action_space["displacement"].low, env.action_space["direction"].low], axis=1)
            space_low = space_low.flatten()
            space_high = np.concatenate([env.action_space["displacement"].high, env.action_space["direction"].high], axis=1)
            space_high = space_high.flatten()
            self.action_space = spaces.Box(low=space_low, high=space_high, shape=(act_dim,), dtype=np.float32)

    def action(self, action):
        # action is a numpy array
        # transform the action from one-dimensional array to dictionary
        if self.is_polar == False:
            return action.reshape([-1,3])
        else:
            action  = action.reshape([-1,3])
            return {
                "displacement": action[:, :1],
                "direction": action[:, 1:],
            }