donkeycar_gym

前提・実現したいこと

donkeycarで強化学習の勉強をしようとしています。
環境設定等を終わらせ、シミュレーターを起動するところまではできました。
gym-donkeycarにddqn.pyというサンプルコードがあるのでこれを実行しようとしてエラーがでました。

環境構築の手順
https://docs.donkeycar.com/guide/install_software/#step-1-install-software-on-host-pc
https://docs.donkeycar.com/guide/simulator/#train

発生している問題・エラーメッセージ

実行文:python ddqn.py --sim /home/<path>/DonkeySimLinux/donkey_sim.x86_64
--simをやりたいのですが、エラーを見ると--env_nameで処理されているのかもしれません。

Traceback (most recent call last):
  File "ddqn.py", line 365, in <module>
    run_ddqn(args)
  File "ddqn.py", line 224, in run_ddqn
    env = gym.make(args.env_name, conf=conf)
  File "/home/user/anaconda3/envs/donkey/lib/python3.7/site-packages/gym/envs/registration.py", line 145, in make
    return registry.make(id, **kwargs)
  File "/home/user/anaconda3/envs/donkey/lib/python3.7/site-packages/gym/envs/registration.py", line 89, in make
    spec = self.spec(path)
  File "/home/user/anaconda3/envs/donkey/lib/python3.7/site-packages/gym/envs/registration.py", line 131, in spec
    raise error.UnregisteredEnv('No registered env with id: {}'.format(id))
gym.error.UnregisteredEnv: No registered env with id: donkey-warehouse-v0

該当のソースコード

condaで環境を作っています。
サイトからymlをダウンロードして、その環境だと[tf2]でしたが、ddqn.pyでは一部[tf1]じゃないと動かない部分があったので、その部分だけ[tensorflow.compat.v1]に書き換えています。(202行目付近)

python
1"""
2file: ddqn.py
3author: Felix Yu
4date: 2018-09-12
5original: https://github.com/flyyufelix/donkey_rl/blob/master/donkey_rl/src/ddqn.py
6"""
7import argparse
8import os
9import random
10import signal
11import sys
12import uuid
13from collections import deque
14
15import cv2
16import gym
17import numpy as np
18import tensorflow as tf
19from tensorflow.keras import backend as K
20from tensorflow.compat.v1.keras import backend as K1
21from tensorflow.keras.layers import Activation, Conv2D, Dense, Flatten
22from tensorflow.keras.models import Sequential
23from tensorflow.keras.optimizers import Adam
24
25#文字数制限に引っかかったので省略しました。
26
27    # only needed if TF==1.13.1
28    # config = tf.ConfigProto()
29    config = tf.compat.v1.ConfigProto()
30    config.gpu_options.allow_growth = True
31    # sess = tf.Session(config=config)
32    sess = tf.compat.v1.Session(config=config)
33    K1.set_session(sess)
34
35    conf = {
36        "exe_path": args.sim,
37        "host": "127.0.0.1",
38        "port": args.port,
39        "body_style": "donkey",
40        "body_rgb": (128, 128, 128),
41        "car_name": "me",
42        "font_size": 100,
43        "racer_name": "DDQN",
44        "country": "USA",
45        "bio": "Learning to drive w DDQN RL",
46        "guid": str(uuid.uuid4()),
47        "max_cte": 10,
48    }
49
50    # Construct gym environment. Starts the simulator if path is given.
51    env = gym.make(args.env_name, conf=conf)
52
53    # not working on windows...
54    def signal_handler(signal, frame):
55        print("catching ctrl+c")
56        env.unwrapped.close()
57        sys.exit(0)
58
59    signal.signal(signal.SIGINT, signal_handler)
60    signal.signal(signal.SIGTERM, signal_handler)
61    signal.signal(signal.SIGABRT, signal_handler)
62
63    # Get size of state and action from environment
64    state_size = (img_rows, img_cols, img_channels)
65    action_space = env.action_space  # Steering and Throttle
66
67    try:
68        agent = DQNAgent(state_size, action_space, train=not args.test)
69
70        throttle = args.throttle  # Set throttle as constant value
71
72        episodes = []
73
74        if os.path.exists(args.model):
75            print("load the saved model")
76            agent.load_model(args.model)
77
78        for e in range(EPISODES):
79
80            print("Episode: ", e)
81
82            done = False
83            obs = env.reset()
84
85            episode_len = 0
86
87            x_t = agent.process_image(obs)
88
89            s_t = np.stack((x_t, x_t, x_t, x_t), axis=2)
90            # In Keras, need to reshape
91            s_t = s_t.reshape(1, s_t.shape[0], s_t.shape[1], s_t.shape[2])  # 1*80*80*4
92
93            while not done:
94
95                # Get action for the current state and go one step in environment
96                steering = agent.get_action(s_t)
97                action = [steering, throttle]
98                next_obs, reward, done, info = env.step(action)
99
100                x_t1 = agent.process_image(next_obs)
101
102                x_t1 = x_t1.reshape(1, x_t1.shape[0], x_t1.shape[1], 1)  # 1x80x80x1
103                s_t1 = np.append(x_t1, s_t[:, :, :, :3], axis=3)  # 1x80x80x4
104
105                # Save the sample <s, a, r, s'> to the replay memory
106                agent.replay_memory(s_t, np.argmax(linear_bin(steering)), reward, s_t1, done)
107                agent.update_epsilon()
108
109                if agent.train:
110                    agent.train_replay()
111
112                s_t = s_t1
113                agent.t = agent.t + 1
114                episode_len = episode_len + 1
115                if agent.t % 30 == 0:
116                    print(
117                        "EPISODE",
118                        e,
119                        "TIMESTEP",
120                        agent.t,
121                        "/ ACTION",
122                        action,
123                        "/ REWARD",
124                        reward,
125                        "/ EPISODE LENGTH",
126                        episode_len,
127                        "/ Q_MAX ",
128                        agent.max_Q,
129                    )
130
131                if done:
132
133                    # Every episode update the target model to be same with model
134                    agent.update_target_model()
135
136                    episodes.append(e)
137
138                    # Save model for each episode
139                    if agent.train:
140                        agent.save_model(args.model)
141
142                    print(
143                        "episode:",
144                        e,
145                        "  memory length:",
146                        len(agent.memory),
147                        "  epsilon:",
148                        agent.epsilon,
149                        " episode length:",
150                        episode_len,
151                    )
152
153    except KeyboardInterrupt:
154        print("stopping run...")
155    finally:
156        env.unwrapped.close()
157
158
159if __name__ == "__main__":
160
161    # Initialize the donkey environment
162    # where env_name one of:
163    env_list = [
164        "donkey-warehouse-v0",
165        "donkey-generated-roads-v0",
166        "donkey-avc-sparkfun-v0",
167        "donkey-generated-track-v0",
168        "donkey-roboracingleague-track-v0",
169        "donkey-waveshare-v0",
170        "donkey-minimonaco-track-v0",
171        "donkey-warren-track-v0",
172        "donkey-thunderhill-track-v0",
173    ]
174
175    parser = argparse.ArgumentParser(description="ddqn")
176    parser.add_argument(
177        "--sim",
178        type=str,
179        default="manual",
180        help="path to unity simulator. maybe be left at manual if you would like to start the sim on your own.",
181    )
182    parser.add_argument("--model", type=str, default="rl_driver.h5", help="path to model")
183    parser.add_argument("--test", action="store_true", help="agent uses learned model to navigate env")
184    parser.add_argument("--port", type=int, default=9091, help="port to use for websockets")
185    parser.add_argument("--throttle", type=float, default=0.3, help="constant throttle for driving")
186    parser.add_argument(
187        "--env_name", type=str, default="donkey-warehouse-v0", help="name of donkey sim environment", choices=env_list
188    )
189
190    args = parser.parse_args()
191
192    run_ddqn(args)

補足情報（FW/ツールのバージョンなど）

[donkey-generated-track-v0]はgym-donkeyにあるようです。
https://gym-donkeycar.readthedocs.io/en/latest/readme.html
また、上記サイトの[Example Usage]も試しましたが、同様のエラーがでました。