An agent, as defined in 2.1 is anything that can perceive its environment through sensors, and act upon that environment through actuators based on its agent program. This can be a dog, robot, or even you. As long as you can perceive the environment and act on it, you are an agent. This notebook will explain how to implement a simple agent, create an environment, and create a program that helps the agent act on the environment based on its percepts.
Before moving on, review the Agent and Environment classes in agents.py.
Let's begin by importing all the functions from the agents.py module and creating our first agent - a blind dog.
In [1]:
from agents import *
class BlindDog(Agent):
def eat(self, thing):
print("Dog: Ate food at {}.".format(self.location))
def drink(self, thing):
print("Dog: Drank water at {}.".format( self.location))
dog = BlindDog()
What we have just done is create a dog who can only feel what's in his location (since he's blind), and can eat or drink. Let's see if he's alive...
In [2]:
print(dog.alive)
A park is an example of an environment because our dog can perceive and act upon it. The Environment class in agents.py is an abstract class, so we will have to create our own subclass from it before we can use it. The abstract class must contain the following methods:
In [3]:
class Food(Thing):
pass
class Water(Thing):
pass
class Park(Environment):
def percept(self, agent):
'''return a list of things that are in our agent's location'''
things = self.list_things_at(agent.location)
return things
def execute_action(self, agent, action):
'''changes the state of the environment based on what the agent does.'''
if action == "move down":
print('{} decided to {} at location: {}'.format(str(agent)[1:-1], action, agent.location))
agent.movedown()
elif action == "eat":
items = self.list_things_at(agent.location, tclass=Food)
if len(items) != 0:
if agent.eat(items[0]): #Have the dog eat the first item
print('{} ate {} at location: {}'
.format(str(agent)[1:-1], str(items[0])[1:-1], agent.location))
self.delete_thing(items[0]) #Delete it from the Park after.
elif action == "drink":
items = self.list_things_at(agent.location, tclass=Water)
if len(items) != 0:
if agent.drink(items[0]): #Have the dog drink the first item
print('{} drank {} at location: {}'
.format(str(agent)[1:-1], str(items[0])[1:-1], agent.location))
self.delete_thing(items[0]) #Delete it from the Park after.
def is_done(self):
'''By default, we're done when we can't find a live agent,
but to prevent killing our cute dog, we will stop before itself - when there is no more food or water'''
no_edibles = not any(isinstance(thing, Food) or isinstance(thing, Water) for thing in self.things)
dead_agents = not any(agent.is_alive() for agent in self.agents)
return dead_agents or no_edibles
In [4]:
class BlindDog(Agent):
location = 1
def movedown(self):
self.location += 1
def eat(self, thing):
'''returns True upon success or False otherwise'''
if isinstance(thing, Food):
return True
return False
def drink(self, thing):
''' returns True upon success or False otherwise'''
if isinstance(thing, Water):
return True
return False
def program(percepts):
'''Returns an action based on it's percepts'''
for p in percepts:
if isinstance(p, Food):
return 'eat'
elif isinstance(p, Water):
return 'drink'
return 'move down'
Let's now run our simulation by creating a park with some food, water, and our dog.
In [5]:
park = Park()
dog = BlindDog(program)
dogfood = Food()
water = Water()
park.add_thing(dog, 1)
park.add_thing(dogfood, 5)
park.add_thing(water, 7)
park.run(5)
Notice that the dog moved from location 1 to 4, over 4 steps, and ate food at location 5 in the 5th step.
Let's continue this simulation for 5 more steps.
In [6]:
park.run(5)
Perfect! Note how the simulation stopped after the dog drank the water - exhausting all the food and water ends our simulation, as we had defined before. Let's add some more water and see if our dog can reach it.
In [7]:
park.add_thing(water, 15)
park.run(10)
This is how to implement an agent, its program, and environment. However, this was a very simple case. Let's try a 2-Dimentional environment now with multiple agents.
To make our Park 2D, we will need to make it a subclass of XYEnvironment instead of Environment. Please note that our park is indexed in the 4th quadrant of the X-Y plane.
We will also eventually add a person to pet the dog.
In [8]:
class Park2D(XYEnvironment):
def percept(self, agent):
'''return a list of things that are in our agent's location'''
things = self.list_things_at(agent.location)
return things
def execute_action(self, agent, action):
'''changes the state of the environment based on what the agent does.'''
if action == "move down":
print('{} decided to {} at location: {}'.format(str(agent)[1:-1], action, agent.location))
agent.movedown()
elif action == "eat":
items = self.list_things_at(agent.location, tclass=Food)
if len(items) != 0:
if agent.eat(items[0]): #Have the dog eat the first item
print('{} ate {} at location: {}'
.format(str(agent)[1:-1], str(items[0])[1:-1], agent.location))
self.delete_thing(items[0]) #Delete it from the Park after.
elif action == "drink":
items = self.list_things_at(agent.location, tclass=Water)
if len(items) != 0:
if agent.drink(items[0]): #Have the dog drink the first item
print('{} drank {} at location: {}'
.format(str(agent)[1:-1], str(items[0])[1:-1], agent.location))
self.delete_thing(items[0]) #Delete it from the Park after.
def is_done(self):
'''By default, we're done when we can't find a live agent,
but to prevent killing our cute dog, we will stop before itself - when there is no more food or water'''
no_edibles = not any(isinstance(thing, Food) or isinstance(thing, Water) for thing in self.things)
dead_agents = not any(agent.is_alive() for agent in self.agents)
return dead_agents or no_edibles
class BlindDog(Agent):
location = [0,1] # change location to a 2d value
direction = Direction("down") # variable to store the direction our dog is facing
def movedown(self):
self.location[1] += 1
def eat(self, thing):
'''returns True upon success or False otherwise'''
if isinstance(thing, Food):
return True
return False
def drink(self, thing):
''' returns True upon success or False otherwise'''
if isinstance(thing, Water):
return True
return False
def program(percepts):
'''Returns an action based on it's percepts'''
for p in percepts:
if isinstance(p, Food):
return 'eat'
elif isinstance(p, Water):
return 'drink'
return 'move down'
Now let's test this new park with our same dog, food and water
In [9]:
park = Park2D(5,20) # park width is set to 5, and height to 20
dog = BlindDog(program)
dogfood = Food()
water = Water()
park.add_thing(dog, [0,1])
park.add_thing(dogfood, [0,5])
park.add_thing(water, [0,7])
morewater = Water()
park.add_thing(morewater, [0,15])
park.run(20)
This works, but our blind dog doesn't make any use of the 2 dimensional space available to him. Let's make our dog more energetic so that he turns and moves forward, instead of always moving down. We'll also need to make appropriate changes to our environment to be able to handle this extra motion.
Let's make our dog turn or move forwards at random - except when he's at the edge of our park - in which case we make him change his direction explicitly by turning to avoid trying to leave the park. Our dog is blind, however, so he wouldn't know which way to turn - he'd just have to try arbitrarily.
| Percept: | Feel Food | Feel Water | Feel Nothing | ||||||
| Action: | eat | drink |
|
In [10]:
from random import choice
turn = False # global variable to remember to turn if our dog hits the boundary
class EnergeticBlindDog(Agent):
location = [0,1]
direction = Direction("down")
def moveforward(self, success=True):
'''moveforward possible only if success (ie valid destination location)'''
global turn
if not success:
turn = True # if edge has been reached, remember to turn
return
if self.direction.direction == Direction.R:
self.location[0] += 1
elif self.direction.direction == Direction.L:
self.location[0] -= 1
elif self.direction.direction == Direction.D:
self.location[1] += 1
elif self.direction.direction == Direction.U:
self.location[1] -= 1
def turn(self, d):
self.direction = self.direction + d
def eat(self, thing):
'''returns True upon success or False otherwise'''
if isinstance(thing, Food):
return True
return False
def drink(self, thing):
''' returns True upon success or False otherwise'''
if isinstance(thing, Water):
return True
return False
def program(percepts):
'''Returns an action based on it's percepts'''
global turn
for p in percepts: # first eat or drink - you're a dog!
if isinstance(p, Food):
return 'eat'
elif isinstance(p, Water):
return 'drink'
if turn: # then recall if you were at an edge and had to turn
turn = False
choice = random.choice((1,2));
else:
choice = random.choice((1,2,3,4)) # 1-right, 2-left, others-forward
if choice == 1:
return 'turnright'
elif choice == 2:
return 'turnleft'
else:
return 'moveforward'
We also need to modify our park accordingly, in order to be able to handle all the new actions our dog wishes to execute. Additionally, we'll need to prevent our dog from moving to locations beyond our park boundary - it just isn't safe for blind dogs to be outside the park by themselves.
In [11]:
class Park2D(XYEnvironment):
def percept(self, agent):
'''return a list of things that are in our agent's location'''
things = self.list_things_at(agent.location)
return things
def execute_action(self, agent, action):
'''changes the state of the environment based on what the agent does.'''
if action == 'turnright':
print('{} decided to {} at location: {}'.format(str(agent)[1:-1], action, agent.location))
agent.turn(Direction.R)
elif action == 'turnleft':
print('{} decided to {} at location: {}'.format(str(agent)[1:-1], action, agent.location))
agent.turn(Direction.L)
elif action == 'moveforward':
loc = copy.deepcopy(agent.location) # find out the target location
if agent.direction.direction == Direction.R:
loc[0] += 1
elif agent.direction.direction == Direction.L:
loc[0] -= 1
elif agent.direction.direction == Direction.D:
loc[1] += 1
elif agent.direction.direction == Direction.U:
loc[1] -= 1
if self.is_inbounds(loc):# move only if the target is a valid location
print('{} decided to move {}wards at location: {}'.format(str(agent)[1:-1], agent.direction.direction, agent.location))
agent.moveforward()
else:
print('{} decided to move {}wards at location: {}, but couldn\'t'.format(str(agent)[1:-1], agent.direction.direction, agent.location))
agent.moveforward(False)
elif action == "eat":
items = self.list_things_at(agent.location, tclass=Food)
if len(items) != 0:
if agent.eat(items[0]):
print('{} ate {} at location: {}'
.format(str(agent)[1:-1], str(items[0])[1:-1], agent.location))
self.delete_thing(items[0])
elif action == "drink":
items = self.list_things_at(agent.location, tclass=Water)
if len(items) != 0:
if agent.drink(items[0]):
print('{} drank {} at location: {}'
.format(str(agent)[1:-1], str(items[0])[1:-1], agent.location))
self.delete_thing(items[0])
def is_done(self):
'''By default, we're done when we can't find a live agent,
but to prevent killing our cute dog, we will stop before itself - when there is no more food or water'''
no_edibles = not any(isinstance(thing, Food) or isinstance(thing, Water) for thing in self.things)
dead_agents = not any(agent.is_alive() for agent in self.agents)
return dead_agents or no_edibles
In [12]:
park = Park2D(3,3)
dog = EnergeticBlindDog(program)
dogfood = Food()
water = Water()
park.add_thing(dog, [0,0])
park.add_thing(dogfood, [1,2])
park.add_thing(water, [2,1])
morewater = Water()
park.add_thing(morewater, [0,2])
print("dog started at [0,0], facing down. Let's see if he found any food or water!")
park.run(20)
This is good, but it still lacks graphics. What if we wanted to visualize our park as it changed? To do that, all we have to do is make our park a subclass of GraphicEnvironment instead of XYEnvironment. Let's see how this looks.
In [13]:
class GraphicPark(GraphicEnvironment):
def percept(self, agent):
'''return a list of things that are in our agent's location'''
things = self.list_things_at(agent.location)
return things
def execute_action(self, agent, action):
'''changes the state of the environment based on what the agent does.'''
if action == 'turnright':
print('{} decided to {} at location: {}'.format(str(agent)[1:-1], action, agent.location))
agent.turn(Direction.R)
elif action == 'turnleft':
print('{} decided to {} at location: {}'.format(str(agent)[1:-1], action, agent.location))
agent.turn(Direction.L)
elif action == 'moveforward':
loc = copy.deepcopy(agent.location) # find out the target location
if agent.direction.direction == Direction.R:
loc[0] += 1
elif agent.direction.direction == Direction.L:
loc[0] -= 1
elif agent.direction.direction == Direction.D:
loc[1] += 1
elif agent.direction.direction == Direction.U:
loc[1] -= 1
if self.is_inbounds(loc):# move only if the target is a valid location
print('{} decided to move {}wards at location: {}'.format(str(agent)[1:-1], agent.direction.direction, agent.location))
agent.moveforward()
else:
print('{} decided to move {}wards at location: {}, but couldn\'t'.format(str(agent)[1:-1], agent.direction.direction, agent.location))
agent.moveforward(False)
elif action == "eat":
items = self.list_things_at(agent.location, tclass=Food)
if len(items) != 0:
if agent.eat(items[0]):
print('{} ate {} at location: {}'
.format(str(agent)[1:-1], str(items[0])[1:-1], agent.location))
self.delete_thing(items[0])
elif action == "drink":
items = self.list_things_at(agent.location, tclass=Water)
if len(items) != 0:
if agent.drink(items[0]):
print('{} drank {} at location: {}'
.format(str(agent)[1:-1], str(items[0])[1:-1], agent.location))
self.delete_thing(items[0])
def is_done(self):
'''By default, we're done when we can't find a live agent,
but to prevent killing our cute dog, we will stop before itself - when there is no more food or water'''
no_edibles = not any(isinstance(thing, Food) or isinstance(thing, Water) for thing in self.things)
dead_agents = not any(agent.is_alive() for agent in self.agents)
return dead_agents or no_edibles
That is the only change we make. The rest of our code stays the same. There is a slight difference in usage though. Every time we create a GraphicPark, we need to define the colors of all the things we plan to put into the park. The colors are defined in typical RGB digital 8-bit format, common across the web.
In [14]:
park = GraphicPark(5,5, color={'EnergeticBlindDog': (200,0,0), 'Water': (0, 200, 200), 'Food': (230, 115, 40)})
dog = EnergeticBlindDog(program)
dogfood = Food()
water = Water()
park.add_thing(dog, [0,0])
park.add_thing(dogfood, [1,2])
park.add_thing(water, [0,1])
morewater = Water()
morefood = Food()
park.add_thing(morewater, [2,4])
park.add_thing(morefood, [4,3])
print("dog started at [0,0], facing down. Let's see if he found any food or water!")
park.run(20)
In [15]:
from ipythonblocks import BlockGrid
from agents import *
color = {"Breeze": (225, 225, 225),
"Pit": (0,0,0),
"Gold": (253, 208, 23),
"Glitter": (253, 208, 23),
"Wumpus": (43, 27, 23),
"Stench": (128, 128, 128),
"Explorer": (0, 0, 255),
"Wall": (44, 53, 57)
}
def program(percepts):
'''Returns an action based on it's percepts'''
print(percepts)
return input()
w = WumpusEnvironment(program, 7, 7)
grid = BlockGrid(w.width, w.height, fill=(123, 234, 123))
def draw_grid(world):
global grid
grid[:] = (123, 234, 123)
for x in range(0, len(world)):
for y in range(0, len(world[x])):
if len(world[x][y]):
grid[y, x] = color[world[x][y][-1].__class__.__name__]
def step():
global grid, w
draw_grid(w.get_world())
grid.show()
w.step()
In [16]:
step()
In [ ]: