add deepracer gym

.gitignore

+__pycache__
\ No newline at end of file

deepracer-gym/README.md

+# Instructions for using the Deepracer Gym Environment
\ No newline at end of file

deepracer-gym/deepracer_gym.egg-info/PKG-INFO

+Metadata-Version: 1.0
+Name: deepracer-gym
+Version: 0.0.1
+Summary: UNKNOWN
+Home-page: UNKNOWN
+Author: UNKNOWN
+Author-email: UNKNOWN
+License: UNKNOWN
+Description: UNKNOWN
+Platform: UNKNOWN

deepracer-gym/deepracer_gym.egg-info/SOURCES.txt

+README.md
+setup.py
+deepracer_gym.egg-info/PKG-INFO
+deepracer_gym.egg-info/SOURCES.txt
+deepracer_gym.egg-info/dependency_links.txt
+deepracer_gym.egg-info/requires.txt
+deepracer_gym.egg-info/top_level.txt
\ No newline at end of file

deepracer-gym/deepracer_gym.egg-info/dependency_links.txt

+

deepracer-gym/deepracer_gym.egg-info/requires.txt

+gym
+zmq
+msgpack
+msgpack_numpy

deepracer-gym/deepracer_gym.egg-info/top_level.txt

+

deepracer-gym/deepracer_gym/__init__.py

+from gym.envs.registration import register
+
+register(id="deepracer-v0", entry_point="deepracer_gym.envs:DeepracerGymEnv")

deepracer-gym/deepracer_gym/envs/__init__.py

+from deepracer_gym.envs.deepracer_gym_env import DeepracerGymEnv
\ No newline at end of file

deepracer-gym/deepracer_gym/envs/deepracer_gym_env.py

+import numpy as np
+import gym
+from deepracer_gym.zmq_client import DeepracerEnvHelper
+
+class DeepracerGymEnv(gym.Env):
+    def __init__(self):
+        self.action_space = gym.spaces.Discrete(5)
+        self.deepracer_helper = DeepracerEnvHelper()
+    
+    def reset(self):
+        observation = self.deepracer_helper.env_reset()
+        return observation
+    
+    def step(self, action):
+        rl_coach_obs = self.deepracer_helper.send_act_rcv_obs(action)
+        observation, reward, done, info = self.deepracer_helper.unpack_rl_coach_obs(rl_coach_obs)
+        return observation, reward, done, info
+
+if __name__ == '__main__':
+    env = DeepracerGymEnv()
+    obs = env.reset()
+    steps_completed = 0
+    episodes_completed = 0
+    total_reward = 0
+    for _ in range(500):
+        observation, reward, done, info = env.step(np.random.randint(5))
+        steps_completed += 1 
+        total_reward += reward
+        if done:
+            episodes_completed += 1
+            print("Episodes Completed:", episodes_completed, "Steps:", steps_completed, "Reward", total_reward)
+            steps_completed = 0
+            total_reward = 0

deepracer-gym/deepracer_gym/zmq_client.py

+import zmq
+import msgpack
+
+import msgpack_numpy as m
+m.patch()
+
+
+class DeepracerZMQClient:
+    def __init__(self, host="127.0.0.1", port=8888):
+        self.host = host
+        self.port = port
+        self.socket = zmq.Context().socket(zmq.REQ)
+        self.socket.set(zmq.SNDTIMEO, 20000)
+        self.socket.set(zmq.RCVTIMEO, 20000)
+        self.socket.connect(f"tcp://{self.host}:{self.port}")
+    
+    def set_agent_ready(self):
+        packed_msg = msgpack.packb({"Agent Ready": 1})
+        self.socket.send(packed_msg)
+
+    def recieve_response(self):
+        packed_response = self.socket.recv()
+        response = msgpack.unpackb(packed_response)
+        return response
+
+    def send_msg(self, msg: dict):
+        packed_msg = msgpack.packb(msg)
+        self.socket.send(packed_msg)
+
+        response = self.recieve_response()
+        return response
+
+class DeepracerEnvHelper:
+    def __init__(self):
+        self.zmq_client = DeepracerZMQClient()
+        self.zmq_client.set_agent_ready()
+        self.obs = None
+        self.previous_done = False
+
+    def send_act_rcv_obs(self, action):
+        action_dict = {"action": action}
+        self.obs = self.zmq_client.send_msg(action_dict)
+        self.previous_done = self.obs['_game_over']
+        return self.obs
+    
+    def env_reset(self):
+        if self.obs is None: # First communication to zmq server
+            self.obs = self.zmq_client.recieve_response()
+        elif self.previous_done: # To prevent dummy episode on already done env
+            pass
+        else: # Can't reset env before episode completes - Passing '1' until episode completes
+            action = 1
+            done = False
+            while not done:
+                self.obs = self.send_act_rcv_obs(action)
+                done = self.obs['_game_over']
+            self.previous_done = True
+
+        return self.obs
+    
+    def unpack_rl_coach_obs(self, rl_coach_obs):
+        observation = rl_coach_obs['_next_state']
+        reward = rl_coach_obs['_reward']
+        done = rl_coach_obs['_game_over']
+        info = rl_coach_obs['info']
+        if type(info) is not dict:
+            info = {}
+        info['goal'] = rl_coach_obs['_goal']
+        return observation, reward, done, info
+
+if __name__ == "__main__":
+    client = DeepracerZMQClient()
+    packed_msg = msgpack.packb({"Ready": 1})
+    client.socket.send(packed_msg)
+    episodes_completed = 0
+    steps_completed = 0
+    while True:
+        packed_response = client.socket.recv()
+        env_response = msgpack.unpackb(packed_response)
+        steps_completed += 1 
+        if env_response['_game_over']:
+            episodes_completed += 1
+            print("Episodes Completed:", episodes_completed, "Steps:", steps_completed)
+            steps_completed = 0
+        packed_action = msgpack.packb({"action": 1})
+        client.socket.send(packed_action)
+
+

deepracer-gym/random_actions_example.py

+import gym
+import numpy as np
+
+import deepracer_gym
+
+env = gym.make('deepracer_gym:deepracer-v0')
+
+obs = env.reset()
+
+print("Deepracer Environment Connected succesfully")
+
+steps_completed = 0
+episodes_completed = 0
+total_reward = 0
+
+for _ in range(500):
+    observation, reward, done, info = env.step(np.random.randint(5))
+  
+    steps_completed += 1 
+    total_reward += reward
+  
+    if done:
+        episodes_completed += 1
+        print("Episodes Completed:", episodes_completed, "Steps:", steps_completed, "Reward", total_reward)
+        steps_completed = 0
+        total_reward = 0
\ No newline at end of file

deepracer-gym/setup.py

+from setuptools import setup
+
+setup(name="deepracer_gym", version="0.0.1", install_requires=["gym","zmq", "msgpack", "msgpack_numpy"])

deepracer-gym/start_deepracer_docker.sh

+docker run -it --rm --name=deepracer -p 8888:8888 -p 5000:5000 --cpus="3" --memory="6g" aicrowd/deepracer:round1_release /bin/bash
\ No newline at end of file

deepracer-gym/stop_deepracer_docker.sh

+docker stop deepracer
\ No newline at end of file