VaccumLand

Gym environment for a trash finding robot

Current supported versions: Python 3.10x

Required Packages: gym

Register Environment With gym

Sample for registering a VacuumLand env with a 20x20 board with 10 trash, a penalty, 400 max steps, and cardinal actions

from vacuum_land import VacuumLand, ActionType
import gym

env_kwargs = {
  'height' : 20, 'width' : 20, 'trash' : 10,
  'as_image' : False, 'penalty' : True, 'seed' : 0,
  'max_steps' : 400, "action_type" : ActionType.CARDINAL,
}
VacuumLand.register(env_kwargs)

Environment Creation

from vacuum_land import VacuumLand, ActionType

env = VacuumLand(height=5, width=5, trash=5, as_image=False, penalty=True, max_steps=25, seed=0)
# VacuumLand.register()
# env = gym.make("VacuumLand-v0")

obs = env.reset()
done = False
while not done:
    action = env.action_space.sample()
    next_obs, reward, done, info = env.step(action)
    env.render()

Current Features

• height
  • Height of the grid
  • default = 5
• width
  • Width of the grid
  • default = 5
• trash
  • How many pieces of trash to randomly place in the grid
  • default = 5
• as_image
  • When True: Returns the observation as (Channel, Height, Width) in PyTorch image format
  • When False: Returns the observation as (Height, Width)
  • default = False
• penalty
  • The penalty for taking a step in the environment that doesn't have a piece of trash on it.
  • When True: Penalty for taking entering a space without trash is -0.01
  • When False: No penalty is applied for taking a step
  • When Numerical Type: Custom value for penalty (Ex. -0.5)
  • default = True
• max_steps
  • The maximum number of steps for the environment
  • When None: Defaults to Height * Width
  • When Int: Custom defined number for max steps (Ex. 400)
  • default = None (Height * Width)
• reward
  • The reward to return when the robot steps over a piece of trash
  • When None: Defaults to 1 / trash
  • When Numerical Type: Custom value for reward (Ex. 1)
• seed
  • The seed for environment creation/reset
  • When None: Randomly seeds the environment when calling np.random.shuffle()
  • When Int: Custom value for seed, gets called during reset in np.random.seed(seed)
  • default = None
• action_type
  • The type of actions you want the environment to use
  • When ActionyType.CARDINAL: Cardinal directions for movement
  • When ActionType.DIAGONAL: Cardinal and diagonal directions for movement
  • default = ActionType.CARDINAL

TODOs

• Get code to follow PEP8 standard
• Convert to a PyPi package to install instead of including the file
• Make it check which version of Python the virtual environment is using, so it can adapt to using different version of Python
• Register the environment on inclusion of the file
• Handle environment with changing starting positions
• Handle having obstacles in the environment (such as holes and walls)
• Have different style of reward (more than just 1 / total trash)
• Multiple styles of action (Cardinal, Diagonal, etc)
• Better environment rendering (PyGame or OpenCV as well as text based)
• Ability to render the environment with a frame rate
• Ability to render the environment for a human to play
• Ability to play a game as human in a better rendered environment
• Make the seed function properly seed the environment (deprecated and handled inside reset)
• Ability to set custom boards
• Modifiers to player/trash value (instead of 1/2 for as_image=False and 122/255 for as_image=True)
• Make a wrapper for registering the env that allows for kwargs