lesson_1_code.py

import os, sys, numpy
module_path = os.path.abspath(os.path.join('../tools'))
if module_path not in sys.path: sys.path.append(module_path)
from DangerousGridWorld import GridWorld


def random_dangerous_grid_world( environment ):
	"""
	Performs a random trajectory on the given Dangerous Grid World environment 
	
	Args:
		environment: OpenAI Gym environment
		
	Returns:
		trajectory: an array containing the sequence of states visited by the agent
	"""
	trajectory = []
	#
	# YOUR CODE HERE!
	#
	for step in range(10):
		#
		# YOUR CODE HERE!
		#
		if False: break # <- Hint: check if the state is terminal
	
	return trajectory


class RecyclingRobot():
	"""
	Class that implements the environment Recycling Robot of the book: 'Reinforcement
	Learning: an introduction, Sutton & Barto'. Example 3.3 page 52 (second edition).
		
	Attributes
	----------
		observation_space : int
			define the number of possible actions of the environment
		action_space: int
			define the number of possible states of the environment
		actions: dict
			a dictionary that translate the 'action code' in human languages
		states: dict
			a dictionary that translate the 'state code' in human languages
		
	Methods
	-------
		reset( self )
			method that reset the environment to an initial state; returns the state
		step( self, action )
			method that perform the action given in input, computes the next state and the reward; returns 
			next_state and reward
		render( self )
			method that print the internal state of the environment
	"""


	def __init__( self ):

		# Loading the default parameters
		self.alfa = 0.7
		self.beta = 0.7
		self.r_search = 0.5
		self.r_wait = 0.2

		# Defining the environment variables
		self.observation_space = None
		self.action_space = None
		self.actions = None
		self.states = None


	def reset( self ):
		#
		# YOUR CODE HERE!
		#
		return self.state


	def step( self, action ):

		reward = 0
		#
		# YOUR CODE HERE!
		#
		return self.state, reward, False, None


	def render( self ):

		#
		# YOUR CODE HERE!
		#
		return True


def main():
	print( "\n************************************************" )
	print( "*  Welcome to the first lesson of the RL-Lab!  *" )
	print( "*             (MDP and Environments)           *" )
	print( "************************************************" )

	print( "\nA) Random Policy on Dangerous Grid World:" )
	env = GridWorld()
	env.render()
	random_trajectory = random_dangerous_grid_world( env )
	print( "\nRandom trajectory generated:", random_trajectory )


	print( "\nB) Custom Environment: Recycling Robot" )
	env = RecyclingRobot()
	state = env.reset()
	ep_reward = 0
	
	for step in range(10):
		a = numpy.random.randint( 0, env.action_space )
		new_state, r, _, _ = env.step( a )
		ep_reward += r
		print( f"\tFrom state '{env.states[state]}' selected action '{env.actions[a]}': \t total reward: {ep_reward:1.1f}" )
		state = new_state


if __name__ == "__main__":
	main()