Azurlane – Food Stalls

A computer-vision bot that automatically plays the Food Stalls memory minigame in Azur Lane, running on an Android emulator on Windows.

The minigame shows a speech bubble with 3 foods in order (the thinking phase); the bubble then disappears and a 3×3 grid of 9 foods appears (the selection phase), where you must click the 3 correct ones, in order. The bot:

1. reads the emulator screen (passive capture via mss);
2. detects the phase from the color level of the bubble and the grid;
3. memorizes the bubble at its sharpest moment;
4. matches each bubble food to a grid cell using ORB keypoints + color + optimal assignment (robust to background, scale and focus);
5. clicks the 3 foods;
6. reads the result screen (⭕ correct / ✗ wrong) to count the points and stops when it reaches the target (default: 25 points), then clicks Quit.

Two modes

File	Mode	Clicks
bot.py	Robot	Exact center of the food, fixed timing. Fast and direct.
modo_humano.py	Human	Randomized position and timing, natural mouse movement.

Both share all of the vision logic; the human mode only changes how it clicks, to reduce the most obvious statistical signs of automation.

Requirements

pip install -r requirements.txt

(opencv-python, numpy, pyautogui, pywin32, mss)

Works with emulators whose window title contains one of: Android Device, BlueStacks, NoxPlayer, LDPlayer, MuMu.

Usage

1. Open the game on the Food Stalls screen, at the lobby with START GAME.

2. Run one of the modes:

   python modo_humano.py    # recommended
   # or
   python bot.py
   

3. During the initial 4 seconds, click the emulator window.

4. Emergency stop: throw the mouse to the top-left corner of the screen.

To check the region framing before running:

python calibrate.py

See TUTORIAL.md for the detailed step-by-step (in Portuguese).

Configuration

Key constants at the top of bot.py:

• POINTS_TARGET – points at which the bot quits (default 25).
• LOBBY_CREDITS_TO_ADD – how many times to click the credits "+1" before START (default 5, the per-batch maximum).
• DEBUG_SAVE – set True to dump debug images for diagnosing.

Human-mode timing/precision ranges live at the top of modo_humano.py.