Pykachu-Volleyball

Introduction

The source code on this repository is an adaption of the code, which is gained by reverse engineering the original game, developed by "(C) SACHI SOFT"

This is a gymnasium environment for single-agent reinforcement learning with a computer as an opponent. Multi-agent environment will be added later, using pettingzoo.

Description

#this is in sample.py
import gymnasium as gym
import pykachu_env

env = gym.make('PykachuVolleyball-v0', 
               render_mode= "human", 
               is_player_2_computer=False)

Create your environment through gym.make('PykachuVolleyball-v0'). Options include render_mode, and is_player_2_computer, which determines which player the computer will control.

Action Space	MultiDiscrete([3, 3, 2])
Observation Space	(432, 304, 3)
Observation High	255
Observation Low	0

Action Space

The action space is MultiDiscrete([3, 3, 2]), and each corresponds to left-right input, up-down input, and power hit input in order.

action_space[0] (left-right movement)

Value	Meaning
0	input left movement
1	NOOP
2	input right movement

action_space[1] (up-down movement)

Value	Meaning
0	input up movement
1	NOOP
2	input down movement

action_space[2] (power hit)

Value	Meaning
0	NOOP
1	input power hit

Observation Space

The observation space is Box(low=0, high=255, shape=(432, 304, 3), dtype=np.uint8). It is the RGB image, displayed to a human player.

#this is in sample.py
for episode in range(5):
    env.reset()

    while True:
        env.render()
        action = env.action_space.sample()
        state, reward, terminated, info = env.step(action)
        if terminated:
            break

env.close()

The step() funcion also returns reward, terminated, and info, along with the above observation.

reward

You'll get +1 when you win, otherwise -1(i.e. if the opponent computer wins).

terminated

True if the ball touches the ground, otherwise False.

info

You can get additional information about the players and tha ball.

{
    "player1": {
        "x": player1.x,
        "y": player1.y,
        "dive_direction" : player1.dive_direction 
    },
    "player2":{
        "x": player2.x,
        "y": player2.y,
        "dive_direction" : player2.dive_direction 
    },
    "ball": {
        "x": ball.x,
        "x_velocity": ball.x_velocity,
        "y": ball.y,
        "y_velocity": ball.y_velocity,
    }
}

TODOs

For the code may not be fully compliant with gymnasium standard APIs, there may be some bugs or issues. Please let me know if there is an error or if you need additional functions. Good luck.

• reset and reward
• refactor to Python convention
• draw background
• show game start message
• transparent Pikachu background
• trail, punch, hyper sprites
• fix computer's improper movements
• ball rotation
• draw shadows
• support render_mode 'rgb_array'
• specific comments
• add sound(and make it optional)