Blackjack Environment

Overview

Blackjack is a card game where players aim to beat the dealer by getting cards that sum closer to 21 without exceeding it. This implementation is based on Example 5.1 in Reinforcement Learning: An Introduction by Sutton and Barto.

Environment Details

Action Space

• Type: Discrete(2)
• Actions:
  • 0: Stick (stop taking cards)
  • 1: Hit (take another card)

Observation Space

• Type: Tuple(Discrete(32), Discrete(11), Discrete(2))
• Components:
  1. Player's current sum
  2. Dealer's showing card (1-10)
  3. Usable ace indicator (0 or 1)

Game Rules

Card Values

• Face cards (Jack, Queen, King): 10 points
• Aces: 11 points (usable) or 1 point
• Number cards (2-9): Face value

Gameplay

1. Initial deal: Dealer gets one face-up and one face-down card; player gets two face-up cards
2. Player's turn: Choose to hit or stick until either sticking or busting
3. Dealer's turn: Reveals face-down card and must hit until sum is 17 or greater

Rewards

• Win: +1
• Lose: -1
• Draw: 0
• Natural Blackjack: +1.5 (if natural=True) or +1 (if natural=False)

Episode Termination

Episodes end when:

• Player busts (sum > 21)
• Player sticks
• Note: Aces are counted as 11 unless doing so would cause a bust

Usage

import gymnasium as gym
env = gym.make("Blackjack-v1")