The source code and dockerfile for the GSW2024 AI Lab.
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
This repo is archived. You can view files and clone it, but cannot push or open issues/pull-requests.
 
 
 
 
 
 

130 lines
4.8 KiB

from __future__ import annotations
from minigrid.core.grid import Grid
from minigrid.core.mission import MissionSpace
from minigrid.core.world_object import Goal, Lava, SlipperyNorth, SlipperyEast, SlipperySouth, SlipperyWest, Ball
from minigrid.minigrid_env import MiniGridEnv, is_slippery
from minigrid.core.tasks import GoTo, DoNothing, PickUpObject, PlaceObject
import numpy as np
class AdversaryEnv(MiniGridEnv):
"""
## Description
"""
def __init__(self, width=7, height=6, generate_wall=True, generate_lava=False, generate_slippery=False ,max_steps: int | None = None, **kwargs):
if max_steps is None:
max_steps = 10 * (width * height)**2
mission_space = MissionSpace(mission_func=self._gen_mission)
self.collision_penalty = -1
super().__init__(
mission_space=mission_space, width=width, height=height, max_steps=max_steps, **kwargs
)
@staticmethod
def _gen_mission():
return "Finish your task while avoiding the adversaries"
def _gen_grid(self, width, height):
self.grid = Grid(width, height)
self.grid.wall_rect(0, 0, width, height)
def step(self, action):
delete_list = list()
for position, box in self.background_tiles.items():
if self.grid.get(*position) is None:
self.grid.set(*position, box)
self.grid.set_background(*position, None)
delete_list.append(tuple(position))
for position in delete_list:
del self.background_tiles[position]
obs, reward, terminated, truncated, info = super().step(action)
agent_pos = self.agent_pos
adv_penalty = 0
if not terminated:
for adversary in self.adversaries.values():
collided = self.move_adversary(adversary, agent_pos)
self.trajectory.append((adversary.color, adversary.adversary_pos, adversary.adversary_dir))
if collided:
terminated = True
info["collision"] = True
try:
reward = self.collision_penalty
except e:
reward = -1
return obs, reward, terminated, truncated, info
def move_adversary(self, adversary, agent_pos):
# fetch current location and forward location
cur_pos = adversary.adversary_pos
current_cell = self.grid.get(*adversary.adversary_pos)
fwd_pos = cur_pos + adversary.dir_vec()
fwd_cell = self.grid.get(*fwd_pos)
collision = False
need_position_update = False
action = adversary.get_action(self)
if action == self.actions.forward and is_slippery(current_cell):
probabilities = current_cell.get_probabilities(adversary.adversary_dir)
possible_fwd_pos, prob = self.get_neighbours_prob(adversary.adversary_pos, probabilities)
fwd_pos_index = np.random.choice(len(possible_fwd_pos), 1, p=prob)
fwd_pos = possible_fwd_pos[fwd_pos_index[0]]
fwd_cell = self.grid.get(*fwd_pos)
need_position_update = True
if action == self.actions.left:
adversary.adversary_dir -= 1
if adversary.adversary_dir < 0:
adversary.adversary_dir += 4
# Rotate right
elif action == self.actions.right:
adversary.adversary_dir = (adversary.adversary_dir + 1) % 4
# Move forward
elif action == self.actions.forward:
if fwd_pos[0] == agent_pos[0] and fwd_pos[1] == agent_pos[1]:
collision = True
if fwd_cell is None or fwd_cell.can_overlap():
adversary.adversary_pos = tuple(fwd_pos)
# Pick up an object
elif action == self.actions.pickup:
if fwd_cell and fwd_cell.can_pickup():
if adversary.carrying is None:
adversary.carrying = fwd_cell
adversary.carrying.cur_pos = np.array([-1, -1])
self.grid.set(fwd_pos[0], fwd_pos[1], None)
# Drop an object
elif action == self.actions.drop:
if not fwd_cell and adversary.carrying:
self.grid.set(fwd_pos[0], fwd_pos[1], adversary.carrying)
adversary.carrying.cur_pos = fwd_pos
adversary.carrying = None
# Toggle/activate an object
elif action == self.actions.toggle:
if fwd_cell:
fwd_cell.toggle(self, fwd_pos)
# Done action (not used by default)
elif action == self.actions.done:
pass
else:
raise ValueError(f"Unknown action: {action}")
if need_position_update and (fwd_cell is None or fwd_cell.can_overlap()):
adversary.adversary_pos = tuple(fwd_pos)
return collision