Reinforcement Learning In Robotics: CartPole environment & Gym toolkit.

Author: DevOpsThinh

README.md

@@ -32,6 +32,13 @@ ___
 2. **[UCI Machine Learning Repository](https://archive.ics.uci.edu/ml/index.php)**
 ___
 
+### Illustrated results
+#### How reinforcement learning helps a robot balance a pole
+- Random movements cannot keep the balance for a long time<br>
+<img src="ml/rl_in_robotics/images/cart-pole-rl.gif" height="480"/> <br>
+- The better than random actions<br>
+<img src="ml/rl_in_robotics/images/cart-pole-balancing-rl.gif" height="480"/> <br>
+
 ### References
 
 1. **[Machine Learning with Python for Everyone](https://www.informit.com/store/machine-learning-with-python-for-everyone-9780134845623)**

ml/rl_in_robotics/cart-pole_balancing_rl.py

@@ -0,0 +1,44 @@
+# Learner: Nguyen Truong Thinh
+# Contact me: [email protected] || +84393280504
+#
+# Topic: Reinforcement Learning (RL): RL in robotics
+#           A classic example: Cart-Pole Reinforcement Learning
+#           (https://www.youtube.com/watch?v=5Q14EjnOJZc)
+
+# Necessary packages
+import random
+from time import sleep
+
+import gym
+
+"""
+    The Gym simulates interaction with the CartPole environment 
+"""
+
+env = gym.make('CartPole-v1')
+
+seed = 1
+random.seed(seed)
+env.seed(seed)
+
+episodes = 10
+
+for i in range(episodes):
+    init_state = env.reset()
+    reward_sum = 0
+
+    while True:
+        env.render()
+        action = 1 if init_state[2] > 0 else 0
+        state, reward, done, debug = env.step(action)
+        reward_sum += reward
+        sleep(.01)
+        if done:
+            print(f'Episode {i} reward: {reward_sum}')
+            sleep(.01)
+            if done:
+                print(f'Episode {i} reward: {reward_sum}')
+                sleep(1)
+                break
+
+env.close()

ml/rl_in_robotics/cart_pole_rl.py

@@ -0,0 +1,47 @@
+# Learner: Nguyen Truong Thinh
+# Contact me: [email protected] || +84393280504
+#
+# Topic: Reinforcement Learning (RL): RL in robotics
+#           A classic example: Cart-Pole Reinforcement Learning
+#           (https://www.youtube.com/watch?v=5Q14EjnOJZc)
+
+# Necessary packages
+import random
+from time import sleep
+
+import gym
+
+"""
+    The Gym simulates interaction with the CartPole environment 
+"""
+
+env = gym.make('CartPole-v1')
+
+seed = 1
+random.seed(seed)
+env.seed(seed)
+
+print(f'Action Space: {env.action_space}')
+print(f'Observation Space: {env.observation_space}')
+
+episodes = 10
+
+for i in range(episodes):
+    init_state = env.reset()
+    reward_sum = 0
+
+    while True:
+        env.render()
+        random_action = random.randint(0, 1)
+        state, reward, done, debug = env.step(random_action)
+        reward_sum += reward
+        sleep(.01)
+        if done:
+            print(f'Episode {i} reward: {reward_sum}')
+            sleep(.01)
+            if done:
+                print(f'Episode {i} reward: {reward_sum}')
+                sleep(1)
+                break
+
+env.close()