parkinglotgym 1.1.2

Gym library for training RL agents on Rush Hour game

ParkingLotGym

A custom AI Gymnasium environment for the Rush hour puzzle: https://en.wikipedia.org/wiki/Rush_Hour_(puzzle).

Library used: .

Example solvable in 60 moves

Random moves are used for this demo. Click on .gif if still.

Example solvable in 5 moves

Random moves are used for this demo. Click if .gif if still.

Usage

Initiating the env via gym

import gymnasium as gym

env_6x6_random = gym.make('Puzzle6x6Moves15-v0')
env_4x4_random = gym.make('Puzzle6x6Moves30-v0')

env_6x6_fixed = gym.make('Puzzle6x6Moves13Fixed-v0')
env_6x6_fixed = gym.make('Puzzle6x6Moves6Fixed-v0')

Initiating the env directly

from parkinglotgym import ParkingLotEnv

env_random = ParkingLotEnv(15)
env_fixed = ParkingLotEnv
("""
    ....O
    FF..O
    .AA..
    ..BB.
    .CC..
    .DD..
""")

Making moves

import gymnasium as gym

env_3x3_fixed = gym.make('Puzzle6x6Moves13Fixed-v0')

# Reset the environment
observation, info = env_3x3_fixed.reset()

# Make a random valid move
import random

# Grab list of vehicles with their available moves.
available_moves = info["available_moves"]
# Drop moves from now, leave only vehicles.
movable_vehicles = [v for v, moves in available_moves.items()]
# Select a random vehicle to move.
vehicle = random.choice(movable_vehicles)
# Select a random move for the chosen vehicle.
move = random.choice(available_moves[vehicle])

# Render the environment. The only render mode is 'human' which renders visual output.
env_3x3_fixed.render()

# Close the environment
env_3x3_fixed.close()

Environment Details

• Action Space: MultiDiscrete([num_vehicles, max_moves*2], start=[2, -max_moves]) - First value selects vehicle ID (starting at 2), second value indicates steps to move (negative for backward, positive for forward). Moving 0 steps is not allowed.
• Observation Space: Box(0, num_vehicles+1, (height, width), int32). Contains values: 0 for empty cells, 1 for walls, 2 and up for vehicles. '2` represents the main vehicle.
• Reward: 0 if the puzzle is solved, -1 if not solved yet, -2 if invalid move.
• Done: True if the puzzle is solved, False otherwise.