Environment

In order to use the Mujoco model for reinforcement learning, it is necessary to create a Python environment of the model.

The environment and therefore the model can be used like this:

import gym
import HexapodEnvironment
import time

env = gym.make('Hexapod-v0')
obs = env.reset()
obs_dim = env.observation_space.shape[0]
act_dim = env.action_space.shape[0]

for i in range(500):
    env.render()
    action = np.array([0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0])
    action = action.reshape((1,-1)).astype(np.float32)
    obs, reward, done, info = env.step(np.squeeze(action, axis=0))
    time.sleep(.2)

Problems with Mujoco

• No collision detection
• Poor documentation -> No tutorials. Examples do not cover all elements (e.g. Sensors).