OpenAI Gym CartPole - Deep Q-Learning (modify reward)

dqn_modify_reward.py

...
next_state, reward, done, info = env.step(action)

# 修改 reward，加快訓練
x, v, theta, omega = next_state
r1 = (env.x_threshold - abs(x)) / env.x_threshold - 0.8 # 小車離中間越近越好
r2 = (env.theta_threshold_radians - abs(theta)) / env.theta_threshold_radians - 0.5 # 柱子越正越好
reward = r1 + r2

dqn.store_transition(state, action, reward, next_state)
...