lzy/sci-gui-agent-benchmark
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Refactor with pyautogui

Browse Source
This commit is contained in:
Timothyxxx
2023-12-02 17:52:00 +08:00
parent 3aebeffec3
commit 992d8f8fce
11 changed files with 144 additions and 515 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
SERVER_SETUP.md
View File

@@ -1,23 +1,6 @@
# Server Setup Guide
- [Linux](#linux)
- [Windows](#windows)
## Linux
<https://averagelinuxuser.com/ssh-into-virtualbox/>
1. `sudo apt install openssh-server`
2. `sudo systemctl enable ssh --now`
3. `sudo ufw disable` (disable firewall - safe for local network, otherwise `sudo ufw allow ssh`)
4. `ip a` - find ip address
5. ssh username@<ip_address>
6. On host, run `ssh-copy-id <username>@<ip_address>`
## Windows
1. Copy and paste the file `windows_server/main.py` to the windows vm
2. Make sure `mouse` and `keyboard` are installed
1. Copy and paste the file `server/main.py` to the windows vm
2. Install the requirements `pip install -r requirements.txt`
3. Run the file `python main.py`
4. `ipconfig /all` and find the ip address

56
desktop_env/controllers/keyboard.py
View File

@@ -1,56 +0,0 @@
from abc import ABC, abstractmethod
from fabric import Connection
from .xdotool import XDoToolController
from .python import PythonController
class AbstractKeyboardController(ABC):
@abstractmethod
def type(self, text: str):
raise NotImplementedError
@abstractmethod
def key(self, key: str):
raise NotImplementedError
@abstractmethod
def key_down(self, key: str):
raise NotImplementedError
@abstractmethod
def key_up(self, key: str):
raise NotImplementedError
class XDoToolKeyboardController(AbstractKeyboardController, XDoToolController):
def __init__(self, ssh_connection: Connection):
super().__init__(ssh_connection=ssh_connection)
def type(self, text: str):
self._execute_xdotool_command(f"type {text}")
def key(self, key: str):
self._execute_xdotool_command(f"key {key}")
def key_down(self, key: str):
self._execute_xdotool_command(f"keydown {key}")
def key_up(self, key: str):
self._execute_xdotool_command(f"keyup {key}")
class PythonKeyboardController(AbstractKeyboardController, PythonController):
def __init__(self, http_server: str):
super().__init__(http_server=http_server)
self.command = "python -c \"import keyboard; {command}\""
def type(self, text: str):
self._execute_python_command(self.command.format(command=f"keyboard.write('{text}')"))
def key(self, key: str):
self._execute_python_command(self.command.format(command=f"keyboard.press_and_release('{key}')"))
def key_down(self, key: str):
self._execute_python_command(self.command.format(command=f"keyboard.press('{key}')"))
def key_up(self, key: str):
self._execute_python_command(self.command.format(command=f"keyboard.release('{key}')"))

162
desktop_env/controllers/mouse.py
View File

@@ -1,162 +0,0 @@
from enum import Enum
from abc import ABC, abstractmethod
from fabric import Connection
import re
from .xdotool import XDoToolController
from .python import PythonController
class MouseClick(Enum):
LEFT = 1
MIDDLE = 2
RIGHT = 3
WHEEL_UP = 4
WHEEL_DOWN = 5
class AbstractMouseController(ABC):
@abstractmethod
def get_mouse(self):
raise NotImplementedError
@abstractmethod
def mouse_move(self, x: int, y: int):
raise NotImplementedError
@abstractmethod
def left_down(self):
raise NotImplementedError
@abstractmethod
def left_up(self):
raise NotImplementedError
@abstractmethod
def left_click(self):
raise NotImplementedError
@abstractmethod
def middle_down(self):
raise NotImplementedError
@abstractmethod
def middle_up(self):
raise NotImplementedError
@abstractmethod
def middle_click(self):
raise NotImplementedError
@abstractmethod
def right_down(self):
raise NotImplementedError
@abstractmethod
def right_up(self):
raise NotImplementedError
@abstractmethod
def right_click(self):
raise NotImplementedError
@abstractmethod
def scroll_up(self):
raise NotImplementedError
@abstractmethod
def scroll_down(self):
raise NotImplementedError
class XDoToolMouseController(AbstractMouseController, XDoToolController):
def __init__(self, ssh_connection: Connection):
super().__init__(ssh_connection=ssh_connection)
def get_mouse(self):
output = self._execute_xdotool_command(f"")
parts = output.split(" ")
x = int(parts[0].split(":")[1])
y = int(parts[1].split(":")[1])
return x, y
def mouse_move(self, x: int, y: int):
self._execute_xdotool_command(f"mousemove {x} {y}")
def left_down(self):
self._execute_xdotool_command(f"mousedown 1")
def left_up(self):
self._execute_xdotool_command(f"mouseup 1")
def left_click(self):
self._execute_xdotool_command(f"click 1")
def middle_down(self):
self._execute_xdotool_command(f"mousedown 2")
def middle_up(self):
self._execute_xdotool_command(f"mouseup 2")
def middle_click(self):
self._execute_xdotool_command(f"click 2")
def right_down(self):
self._execute_xdotool_command(f"mousedown 3")
def right_up(self):
self._execute_xdotool_command(f"mouseup 3")
def right_click(self):
self._execute_xdotool_command(f"click 3")
def scroll_up(self):
self._execute_xdotool_command(f"click 4")
def scroll_down(self):
self._execute_xdotool_command(f"click 5")
class PythonMouseController(AbstractMouseController, PythonController):
def __init__(self, http_server: str):
super().__init__(http_server=http_server)
self.command = "py -c \"import mouse; {command}\""
def get_mouse(self):
response = self._execute_python_command(self.command.format(command=f"print(mouse.get_position())"))
numbers = re.findall(r'-?\d+', response["output"])
x, y = map(int, numbers)
return x, y
def mouse_move(self, x: int, y: int):
self._execute_python_command(self.command.format(command=f"mouse.move({x}, {y})"))
def left_down(self):
self._execute_python_command(self.command.format(command="mouse.press(button='left')"))
def left_up(self):
self._execute_python_command(self.command.format(command="mouse.release(button='left')"))
def left_click(self):
self._execute_python_command(self.command.format(command="mouse.click(button='left')"))
def middle_down(self):
self._execute_python_command(self.command.format(command="mouse.press(button='middle')"))
def middle_up(self):
self._execute_python_command(self.command.format(command="mouse.release(button='middle')"))
def middle_click(self):
self._execute_python_command(self.command.format(command="mouse.click(button='middle')"))
def right_down(self):
self._execute_python_command(self.command.format(command="mouse.press(button='right')"))
def right_up(self):
self._execute_python_command(self.command.format(command="mouse.release(button='right')"))
def right_click(self):
self._execute_python_command(self.command.format(command="mouse.click(button='right')"))
def scroll_up(self):
self._execute_python_command(self.command.format(command="mouse.wheel(10)"))
def scroll_down(self):
self._execute_python_command(self.command.format(command="mouse.wheel(-10)"))

40
desktop_env/controllers/python.py
View File

@@ -1,15 +1,31 @@
import requests
import json
import requests
class PythonController:
def __init__(self, http_server: str):
def __init__(self, http_server: str, pkgs_prefix: str = "py -c \"import pyautogui; {command}\""):
self.http_server = http_server
self.pkgs_prefix = pkgs_prefix # fixme: this is a hacky way to execute python commands. fix it and combine it with installation of packages
def _execute_python_command(self, command: str) -> None:
payload = json.dumps({
"command": command
})
def get_screenshot(self):
"""
Gets a screenshot from the server. With the cursor.
"""
response = requests.get(self.http_server + "/screenshot")
if response.status_code == 200:
return response.content
else:
print("Failed to get screenshot. Status code:", response.status_code)
return None
def execute_python_command(self, command: str) -> None:
"""
Executes a python command on the server.
It can be used to execute the pyautogui commands, or... any other python command. who knows?
"""
command = self.pkgs_prefix.format(command=command)
payload = json.dumps({"command": command})
headers = {
'Content-Type': 'application/json'
}
@@ -23,15 +39,3 @@ class PythonController:
return response.json()
except requests.exceptions.RequestException as e:
print("An error occurred while trying to execute the command:", e)
# example usage
if __name__ == '__main__':
# replace with your actual server URL of the vm
server_url = "http://192.168.7.129:5000"
controller = PythonController(server_url)
# example commands
python_command = "python -c \"import keyboard; keyboard.write('hello world')\""
python_command = "python -c \"import mouse; mouse.move(100,100);mouse.right_click()\""
controller._execute_python_command(python_command)

11
desktop_env/controllers/xdotool.py
View File

@@ -1,11 +0,0 @@
from fabric import Connection
from typing import List
class XDoToolController:
def __init__(self, ssh_connection: Connection):
self.ssh_connection = ssh_connection
def _execute_xdotool_command(self, command: List[str]) -> None:
result = self.ssh_connection.run(f"DISPLAY=:0 xdotool {command}", hide=True)
return result.stdout.strip()

194
desktop_env/envs/desktop_env.py
View File

@@ -1,34 +1,20 @@
from __future__ import annotations
import os
from enum import Enum
from typing import Literal, List, Tuple
import subprocess
from fabric import Connection
import time
import uuid
from typing import List
import gymnasium as gym
from gymnasium import spaces
import numpy as np
import uuid
from PIL import Image
from desktop_env.controllers.mouse import MouseClick, AbstractMouseController, XDoToolMouseController, \
PythonMouseController
from desktop_env.controllers.keyboard import AbstractKeyboardController, XDoToolKeyboardController, \
PythonKeyboardController
from desktop_env.controllers.python import PythonController
class Action(Enum):
CLICK = 0
MOUSE_DOWN = 1
MOUSE_UP = 2
MOUSE_MOVE = 3
KEY = 4
KEY_DOWN = 5
KEY_UP = 6
TYPE = 7
VM_TYPE = Literal['ubuntu', 'windows']
def _execute_command(command: List[str]) -> None:
result = subprocess.run(command, stdout=subprocess.PIPE, stderr=subprocess.PIPE, timeout=60, text=True)
if result.returncode != 0:
raise Exception("\033[91m" + result.stdout + result.stderr + "\033[0m")
class DesktopEnv(gym.Env):
@@ -37,67 +23,20 @@ class DesktopEnv(gym.Env):
def __init__(
self,
path_to_vm: str,
username: str,
password: str = None,
host: str = "192.168.7.128:5000",
snapshot_path: str = "base",
vm_os: VM_TYPE = "ubuntu"
):
# The path to the vmx file of your vm
# Initialize environment variables
self.path_to_vm = path_to_vm
# username and password for your vm
self.username = username
self.password = password
self.host = host
self.snapshot_path = snapshot_path # todo: handling the logic of snapshot directory
# Initialize emulator
# Initialize emulator and controller
print("Initializing...")
self._start_emulator()
self.controller = PythonController(http_server=self.host)
# set up controllers
self.mouse_controller, self.keyboard_controller = self._create_controllers(vm_os)
# Get the screen size
self.screen_width, self.screen_height = self._get_screensize()
# Define the action and observation space
self.action_space = spaces.Dict({
"action_type": spaces.Discrete(len(Action)),
"click_type": spaces.Discrete(len(MouseClick)),
"x": spaces.Discrete(self.screen_width),
"y": spaces.Discrete(self.screen_height),
"key": spaces.MultiDiscrete([128] * 10), # max 10 characters, ASCII
"text": spaces.MultiDiscrete([128] * 10) # max 10 characters, ASCII
})
self.observation_space = spaces.Box(low=0, high=255, shape=(self.screen_width, self.screen_height, 3),
dtype=np.uint8)
# Additional setup
self.metadata = {'render.modes': ['rgb_array']}
def _get_screensize(self):
screenshot_path = self._get_obs()
img = Image.open(screenshot_path)
return img.size
def _create_controllers(self, vm_os: VM_TYPE) -> Tuple[AbstractMouseController, AbstractKeyboardController]:
if vm_os == "ubuntu":
ssh_connection = Connection(host=self.host, user=self.username, connect_kwargs={"password": self.password})
mouse_controller = XDoToolMouseController(ssh_connection)
keyboard_controller = XDoToolKeyboardController(ssh_connection)
elif vm_os == "windows":
mouse_controller = PythonMouseController(http_server=self.host)
keyboard_controller = PythonKeyboardController(http_server=self.host)
else:
raise NotImplementedError(vm_os)
return mouse_controller, keyboard_controller
# todo: define the action space and the observation space as gym did
def _start_emulator(self):
while True:
@@ -109,52 +48,35 @@ class DesktopEnv(gym.Env):
break
else:
print("Starting VM...")
self._execute_command(["vmrun", "-T", "ws", "start", self.path_to_vm])
_execute_command(["vmrun", "-T", "ws", "start", self.path_to_vm])
time.sleep(10)
except subprocess.CalledProcessError as e:
print(f"Error executing command: {e.output.decode().strip()}")
def _execute_command(self, command: List[str]) -> None:
result = subprocess.run(command, stdout=subprocess.PIPE, stderr=subprocess.PIPE, timeout=60, text=True)
if result.returncode != 0:
raise Exception("\033[91m" + result.stdout + result.stderr + "\033[0m")
def _save_state(self):
self._execute_command(["vmrun", "-T", "ws" "snapshot", self.path_to_vm, self.snapshot_path])
_execute_command(["vmrun", "-T", "ws" "snapshot", self.path_to_vm, self.snapshot_path])
def _get_screenshot(self):
random_uuid = str(uuid.uuid4())
os.makedirs(os.path.join("tmp", random_uuid), exist_ok=True)
image_path = os.path.join("tmp", random_uuid, "screenshot.png")
if self.password:
self._execute_command(
["vmrun", "-T", "ws", "-gu", self.username, "-gp", self.password, "captureScreen", self.path_to_vm, image_path])
else:
self._execute_command(
["vmrun", "-T", "ws", "-gu", self.username, "captureScreen", self.path_to_vm, image_path])
# Get the screenshot and save to the image_path
screenshot = self.controller.get_screenshot()
with open(image_path, "wb") as f:
f.write(screenshot)
return image_path
def _get_obs(self):
screenshot_image_path = self._get_screenshot()
self._add_cursor(screenshot_image_path)
return screenshot_image_path
def _add_cursor(self, img_path: str):
x, y = self.mouse_controller.get_mouse()
cursor_image = Image.open("./desktop_env/assets/cursor.png")
cursor_image = cursor_image.resize((int(cursor_image.width / 2), int(cursor_image.height / 2)))
screenshot = Image.open(img_path)
screenshot.paste(cursor_image, (x, y), cursor_image)
screenshot.save(img_path)
def reset(self):
def reset(self, seed=None, options=None):
print("Resetting environment...")
print("Reverting to snapshot to {}...".format(self.snapshot_path))
self._execute_command(["vmrun", "-T", "ws", "revertToSnapshot", self.path_to_vm, self.snapshot_path])
_execute_command(["vmrun", "-T", "ws", "revertToSnapshot", self.path_to_vm, self.snapshot_path])
time.sleep(5)
print("Starting emulator...")
@@ -165,75 +87,11 @@ class DesktopEnv(gym.Env):
return observation
def step(self, action):
if isinstance(action, list):
for a in action:
observation, reward, done, info = self.step(a)
return observation, reward, done, info
# todo: handle the case when the action is not a single action
try:
action_type = Action(action['action_type'])
except KeyError:
done = True
return self._get_obs(), 0, done, {}
if action_type == Action.CLICK:
click = MouseClick(action['click_type'])
if click == MouseClick.LEFT:
self.mouse_controller.left_click()
elif click == MouseClick.MIDDLE:
self.mouse_controller.middle_click()
elif click == MouseClick.RIGHT:
self.mouse_controller.right_click()
elif click == MouseClick.WHEEL_UP:
self.mouse_controller.scroll_up()
elif click == MouseClick.WHEEL_DOWN:
self.mouse_controller.scroll_down()
elif action_type == Action.MOUSE_DOWN:
click = MouseClick(action['click_type'])
if click == MouseClick.LEFT:
self.mouse_controller.left_down()
elif click == MouseClick.MIDDLE:
self.mouse_controller.middle_down()
elif click == MouseClick.RIGHT:
self.mouse_controller.right_down()
elif click == MouseClick.WHEEL_UP:
self.mouse_controller.scroll_up()
elif click == MouseClick.WHEEL_DOWN:
self.mouse_controller.scroll_down()
elif action_type == Action.MOUSE_UP:
click = MouseClick(action['click_type'])
if click == MouseClick.LEFT:
self.mouse_controller.left_up()
elif click == MouseClick.MIDDLE:
self.mouse_controller.middle_up()
elif click == MouseClick.RIGHT:
self.mouse_controller.right_up()
elif click == MouseClick.WHEEL_UP:
self.mouse_controller.scroll_up()
elif click == MouseClick.WHEEL_DOWN:
self.mouse_controller.scroll_down()
elif action_type == Action.MOUSE_MOVE:
self.mouse_controller.mouse_move(x=action['x'], y=action['y'])
elif action_type == Action.KEY:
self.keyboard_controller.key(action['key'])
elif action_type == Action.KEY_DOWN:
self.keyboard_controller.key_down(action['key'])
elif action_type == Action.KEY_UP:
self.keyboard_controller.key_up(action['key'])
elif action_type == Action.TYPE:
for key in action['text']:
if key == "\r" or key == "\n":
self.keyboard_controller.key("enter")
else:
self.keyboard_controller.key(key)
# sleep for 0.05 seconds with some random noise
time.sleep(0.05 + np.random.normal(0, 0.01))
# Capture new state
# Our action space is the set of all possible python commands insides `pyautogui`
self.controller.execute_python_command(action)
observation = self._get_obs()
reward = 0 # Define reward calculation
done = False # Define episode termination condition
reward = 0 # todo: Define reward calculation for each example
done = False # todo: Define episode termination condition for each example
info = {}
return observation, reward, done, info
@@ -244,4 +102,4 @@ class DesktopEnv(gym.Env):
raise ValueError('Unsupported render mode: {}'.format(mode))
def close(self):
self._execute_command(["vmrun", "stop", self.path_to_vm])
_execute_command(["vmrun", "stop", self.path_to_vm])

64
desktop_env/server/main.py Normal file
View File

@@ -0,0 +1,64 @@
import os
import platform
import subprocess
import Xlib.display
import pyautogui
from PIL import ImageGrab
from flask import Flask, request, jsonify, send_file
app = Flask(__name__)
pyautogui.PAUSE = 0
pyautogui.DARWIN_CATCH_UP_TIME = 0
@app.route('/execute', methods=['POST'])
def execute_command():
data = request.json
# The 'command' key in the JSON request should contain the command to be executed.
command = data.get('command', '')
# Execute the command without any safety checks.
try:
result = subprocess.run(command, shell=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True)
return jsonify({
'status': 'success',
'output': result.stdout,
'error': result.stderr
})
except Exception as e:
return jsonify({
'status': 'error',
'message': str(e)
}), 500
@app.route('/screenshot', methods=['GET'])
def capture_screen_with_cursor():
file_path = os.path.join("screenshots", "screenshot.png")
user_platform = platform.system()
# Ensure the screenshots directory exists
os.makedirs(os.path.dirname(file_path), exist_ok=True)
if user_platform == "Windows":
screenshot = pyautogui.screenshot()
screenshot.save(file_path)
elif user_platform == "Linux":
# Use xlib to prevent scrot dependency for Linux
screen = Xlib.display.Display().screen()
size = screen.width_in_pixels, screen.height_in_pixels
screenshot = ImageGrab.grab(bbox=(0, 0, size[0], size[1]))
screenshot.save(file_path)
elif user_platform == "Darwin": # (Mac OS)
# Use the screencapture utility to capture the screen with the cursor
subprocess.run(["screencapture", "-C", file_path])
else:
print(f"The platform you're using ({user_platform}) is not currently supported")
return send_file(file_path, mimetype='image/png')
if __name__ == '__main__':
app.run(debug=True, host="0.0.0.0")

27
desktop_env/windows_server/main.py
View File

@@ -1,27 +0,0 @@
from flask import Flask, request, jsonify
import subprocess
app = Flask(__name__)
@app.route('/execute', methods=['POST'])
def execute_command():
data = request.json
# The 'command' key in the JSON request should contain the command to be executed.
command = data.get('command', '')
# Execute the command without any safety checks.
try:
result = subprocess.run(command, shell=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True)
return jsonify({
'status': 'success',
'output': result.stdout,
'error': result.stderr
})
except Exception as e:
return jsonify({
'status': 'error',
'message': str(e)
}), 500
if __name__ == '__main__':
app.run(debug=True, host="0.0.0.0")

19
gpt_4v_agent_exp.py
View File

@@ -8,29 +8,26 @@ import uuid
def gpt_4v_agent():
api_key = os.environ.get("OPENAI_API_KEY")
# meta_info = {
# "instruction": "Open WSJ website to get latest news",
# "task_name": "open_wsj",
# "snapshot_path": "base",
# }
meta_info = {
"instruction": "Clear the recycle bin",
"task_name": "clean_recycle_bin",
"instruction": "Open WSJ website to get latest news",
"task_name": "open_wsj",
"snapshot_path": "base",
}
# meta_info = {
# "instruction": "Clear the recycle bin",
# "task_name": "clean_recycle_bin",
# "snapshot_path": "base",
# }
agent = GPT4v_Agent(api_key=api_key, instruction=meta_info["instruction"])
env = DesktopEnv(
path_to_vm=r"""C:\Users\tianbaox\Documents\Virtual Machines\Win10\Win10.vmx""",
# automitically load the snapshot and start the vm
# path_to_vm="/home/yuri/vmware/Ubuntu 64-bit/Ubuntu 64-bit.vmx",
snapshot_path="base",
username="tianbaox",
password="951753",
# host="192.168.7.128",
host="http://192.168.13.128:5000",
vm_os="windows"
)
# reset the environment to certain snapshot

63
main.py
View File

@@ -1,55 +1,31 @@
from pprint import pprint
from desktop_env.envs.desktop_env import DesktopEnv, Action, MouseClick
def get_human_action():
"""
Prompts the human player for an action and returns a structured action.
"""
print("\nAvailable actions:", [action.name for action in Action])
action_type = None
while action_type not in [action.value for action in Action]:
action_type = Action[input("Enter the type of action: ".strip())].value
action = {"action_type": action_type}
if action_type == Action.CLICK.value or action_type == Action.MOUSE_DOWN.value or action_type == Action.MOUSE_UP.value:
print("\n Available clicks:", [action.name for action in MouseClick])
click_type = input("Enter click type: ")
action["click_type"] = MouseClick[click_type].value
if action_type == Action.MOUSE_MOVE.value:
x = int(input("Enter x-coordinate for mouse move: "))
y = int(input("Enter y-coordinate for mouse move: "))
action["x"] = x
action["y"] = y
if action_type == Action.KEY.value:
key = input("Enter the key to press: ")
action["key"] = [ord(c) for c in key]
if action_type == Action.TYPE.value:
text = input("Enter the text to type: ")
action["text"] = [ord(c) for c in text]
return action
from desktop_env.envs.desktop_env import DesktopEnv
def human_agent():
"""
Runs the Gym environment with human input.
"""
env = DesktopEnv(path_to_vm="/home/yuri/vmware/Windows 10 x64/Windows 10 x64.vmx",
# path_to_vm="/home/yuri/vmware/Ubuntu 64-bit/Ubuntu 64-bit.vmx",
username="user",
password="password",
# host="192.168.7.128",
host="http://192.168.7.129:5000",
vm_os="windows")
observation = env.reset()
env = DesktopEnv(
path_to_vm=r"""C:\Users\tianbaox\Documents\Virtual Machines\Win10\Win10.vmx""",
# path_to_vm="/home/yuri/vmware/Ubuntu 64-bit/Ubuntu 64-bit.vmx",
# host="192.168.7.128",
host="http://192.168.13.128:5000",
)
# reset the environment to certain snapshot
# observation = env.reset()
done = False
while not done:
action = get_human_action()
# action = get_human_action()
# action = {
# "action_type": 0,
# "click_type": 3,
# }
action = "pyautogui.dragTo(100, 200, button='left')"
observation, reward, done, info = env.step(action)
print("Observation:", observation)
print("Reward:", reward)
@@ -64,5 +40,6 @@ def human_agent():
env.close()
print("Environment closed.")
if __name__ == "__main__":
human_agent()

2
mm_agents/gpt_4v_agent.py
View File

@@ -88,6 +88,8 @@ class GPT4v_Agent:
traj_to_show = []
for i in range(len(self.trajectory)):
traj_to_show.append(self.trajectory[i]["content"][0]["text"])
if len(self.trajectory[i]["content"]) > 1:
traj_to_show.append("screenshot_obs")
print("Trajectory:", traj_to_show)
payload = {
"model": self.model,