def get_next_states(self, state, action):
"""Returns the next state given a state and an action."""
action = int(action)
orientation, x, y = state.agent_pos
new_orientation, new_x, new_y = state.agent_pos
new_tree_states = deepcopy(state.tree_states)
new_bucket_states = deepcopy(state.bucket_states)
new_carrying_apple = state.carrying_apple
if action == Direction.get_number_from_direction(Direction.STAY):
pass
elif action < len(Direction.ALL_DIRECTIONS):
new_orientation = action
move_x, move_y = Direction.move_in_direction_number((x, y), action)
# New position is legal
if (0 <= move_x < self.width and 0 <= move_y < self.height
and (move_x, move_y) in self.possible_agent_locations):
new_x, new_y = move_x, move_y
else:
# Move only changes orientation, which we already handled
pass
elif action == 5:
obj_pos = Direction.move_in_direction_number((x, y), orientation)
if state.carrying_apple:
# We always drop the apple
new_carrying_apple = False
# If we're facing a bucket, it goes there
if obj_pos in new_bucket_states:
prev_apples = new_bucket_states[obj_pos]
new_bucket_states[obj_pos] = min(prev_apples + 1,
self.bucket_capacity)
elif obj_pos in new_tree_states and new_tree_states[obj_pos]:
new_carrying_apple = True
new_tree_states[obj_pos] = False
else:
# Interact while holding nothing and not facing a tree.
pass
else:
raise ValueError("Invalid action {}".format(action))
new_pos = new_orientation, new_x, new_y
def make_state(prob_apples_tuple):
prob, tree_apples = prob_apples_tuple
trees = dict(zip(self.tree_locations, tree_apples))
s = ApplesState(new_pos, trees, new_bucket_states,
new_carrying_apple)
return (prob, s, 0)
# For apple regeneration, don't regenerate apples that were just picked,
# so use the apple booleans from the original state
old_tree_apples = [
state.tree_states[loc] for loc in self.tree_locations
]
new_tree_apples = [new_tree_states[loc] for loc in self.tree_locations]
return list(
map(make_state, self.regen_apples(old_tree_apples,
new_tree_apples)))
def get_next_state(self, state, action):
"""Returns the next state given a state and an action."""
action = int(action)
new_x, new_y = Direction.move_in_direction_number(
state.agent_pos, action)
# New position is still in bounds:
if not (0 <= new_x < self.width and 0 <= new_y < self.height):
new_x, new_y = state.agent_pos
new_agent_pos = new_x, new_y
new_train_pos, new_train_life = state.train_pos, state.train_life
new_battery_present = deepcopy(state.battery_present)
new_carrying_battery = state.carrying_battery
if new_agent_pos == state.train_pos and state.carrying_battery:
new_train_life = 10
new_carrying_battery = False
if new_train_life > 0:
new_train_pos = self.train_transition[state.train_pos]
new_train_life -= 1
if (new_agent_pos in state.battery_present
and state.battery_present[new_agent_pos]
and not state.carrying_battery):
new_carrying_battery = True
new_battery_present[new_agent_pos] = False
result = BatteriesState(
new_agent_pos,
new_train_pos,
new_train_life,
new_battery_present,
new_carrying_battery,
)
return result
def get_next_state(self, state, action):
"""Returns the next state given a state and an action."""
action = int(action)
new_x, new_y = Direction.move_in_direction_number(
state.agent_pos, action)
# New position is still in bounds:
if not (0 <= new_x < self.width and 0 <= new_y < self.height):
new_x, new_y = state.agent_pos
new_agent_pos = new_x, new_y
new_vase_states = deepcopy(state.vase_states)
if new_agent_pos in new_vase_states:
new_vase_states[new_agent_pos] = False # Break the vase
return RoomState(new_agent_pos, new_vase_states)
def get_next_state(self, state, action):
"""Returns the next state given a state and an action."""
action = int(action)
if action == Direction.get_number_from_direction(Direction.STAY):
pass
elif action < len(Direction.ALL_DIRECTIONS):
move_x, move_y = Direction.move_in_direction_number(state, action)
# New position is legal
if 0 <= move_x < self.width and 0 <= move_y < self.height:
state = move_x, move_y
else:
# Move only changes orientation, which we already handled
pass
else:
raise ValueError("Invalid action {}".format(action))
return state
def get_next_state(self, state, action):
"""Return the next state given a state and an action."""
action = int(action)
new_x, new_y = Direction.move_in_direction_number(
state.agent_pos, action)
# New position is still in bounds:
if not (0 <= new_x < self.width and 0 <= new_y < self.height):
new_x, new_y = state.agent_pos
new_agent_pos = new_x, new_y
new_vase_states = deepcopy(state.vase_states)
new_train_pos, new_train_intact = state.train_pos, state.train_intact
if state.train_intact:
new_train_pos = self.train_transition[state.train_pos]
# Break the vase and train if appropriate
if new_agent_pos in new_vase_states:
new_vase_states[new_agent_pos] = False
if new_agent_pos == new_train_pos:
new_train_intact = False
return TrainState(new_agent_pos, new_vase_states, new_train_pos,
new_train_intact)