/
OS-World3d32556
import ast
import base64
import logging
import math
import re
import xml.etree.ElementTree as ET
from io import BytesIO
from typing import Dict, List
import os
import backoff
import numpy as np
from PIL import Image
from requests.exceptions import SSLError
import openai
from openai import OpenAI
from google.api_core.exceptions import (
BadRequest,
InternalServerError,
InvalidArgument,
ResourceExhausted,
)
from mm_agents.accessibility_tree_wrap.heuristic_retrieve import (
filter_nodes,
)
from mm_agents.prompts import (
UITARS_ACTION_SPACE,
UITARS_CALL_USR_ACTION_SPACE,
UITARS_USR_PROMPT_NOTHOUGHT,
UITARS_USR_PROMPT_THOUGHT,
)
from loguru import logger
FINISH_WORD = "finished"
WAIT_WORD = "wait"
ENV_FAIL_WORD = "error_env"
CALL_USER = "call_user"
pure_text_settings = ["a11y_tree"]
attributes_ns_ubuntu = "https://accessibility.windows.example.org/ns/attributes"
attributes_ns_windows = "https://accessibility.windows.example.org/ns/attributes"
state_ns_ubuntu = "https://accessibility.ubuntu.example.org/ns/state"
state_ns_windows = "https://accessibility.windows.example.org/ns/state"
component_ns_ubuntu = "https://accessibility.ubuntu.example.org/ns/component"
component_ns_windows = "https://accessibility.windows.example.org/ns/component"
value_ns_ubuntu = "https://accessibility.ubuntu.example.org/ns/value"
value_ns_windows = "https://accessibility.windows.example.org/ns/value"
class_ns_windows = "https://accessibility.windows.example.org/ns/class"
# More namespaces defined in OSWorld, please check desktop_env/server/main.py
# 定义一个函数来解析每个 action
def parse_action(action_str):
try:
# 解析字符串为 AST 节点
node = ast.parse(action_str, mode='eval')
# 确保节点是一个表达式
if not isinstance(node, ast.Expression):
raise ValueError("Not an expression")
# 获取表达式的主体
call = node.body
# 确保主体是一个函数调用
if not isinstance(call, ast.Call):
raise ValueError("Not a function call")
# 获取函数名
if isinstance(call.func, ast.Name):
func_name = call.func.id
elif isinstance(call.func, ast.Attribute):
func_name = call.func.attr
else:
func_name = None
# 获取关键字参数
kwargs = {}
for kw in call.keywords:
key = kw.arg
# 处理不同类型的值,这里假设都是常量
if isinstance(kw.value, ast.Constant):
value = kw.value.value
elif isinstance(kw.value, ast.Str): # 兼容旧版本 Python
value = kw.value.s
else:
value = None
kwargs[key] = value
return {
'function': func_name,
'args': kwargs
}
except Exception as e:
print(f"Failed to parse action '{action_str}': {e}")
return None
def escape_single_quotes(text):
# 匹配未转义的单引号(不匹配 \\')
pattern = r"(?<!\\)'"
return re.sub(pattern, r"\\'", text)
def parse_action_qwen2vl(text, factor, image_height, image_width):
text = text.strip()
# 正则表达式匹配 Action 字符串
if text.startswith("Thought:"):
thought_pattern = r"Thought: (.+?)(?=\s*Action:|$)"
thought_hint = "Thought: "
elif text.startswith("Reflection:"):
thought_pattern = r"Reflection: (.+?)Action_Summary: (.+?)(?=\s*Action:|$)"
thought_hint = "Reflection: "
elif text.startswith("Action_Summary:"):
thought_pattern = r"Action_Summary: (.+?)(?=\s*Action:|$)"
thought_hint = "Action_Summary: "
else:
# 修复:当没有明确的"Thought:"标识时,提取Action:之前的所有内容作为思考
thought_pattern = r"(.+?)(?=\s*Action:|$)"
thought_hint = ""
reflection, thought = None, None
thought_match = re.search(thought_pattern, text, re.DOTALL)
if thought_match:
if len(thought_match.groups()) == 1:
thought = thought_match.group(1).strip()
elif len(thought_match.groups()) == 2:
thought = thought_match.group(2).strip()
reflection = thought_match.group(1).strip()
assert "Action:" in text
action_str = text.split("Action:")[-1]
tmp_all_action = action_str.split("\n\n")
all_action = []
for action_str in tmp_all_action:
if "type(content" in action_str:
# 正则表达式匹配 content 中的字符串并转义单引号
def escape_quotes(match):
content = match.group(1) # 获取 content 的值
return content
# 使用正则表达式进行替换
pattern = r"type\(content='(.*?)'\)" # 匹配 type(content='...')
content = re.sub(pattern, escape_quotes, action_str)
# 处理字符串
action_str = escape_single_quotes(content)
action_str = "type(content='" + action_str + "')"
all_action.append(action_str)
parsed_actions = [parse_action(action.replace("\n","\\n").lstrip()) for action in all_action]
actions = []
for action_instance, raw_str in zip(parsed_actions, all_action):
if action_instance == None:
print(f"Action can't parse: {raw_str}")
continue
action_type = action_instance["function"]
params = action_instance["args"]
# import pdb; pdb.set_trace()
action_inputs = {}
for param_name, param in params.items():
if param == "": continue
param = param.lstrip() # 去掉引号和多余的空格
# 处理start_box或者end_box参数格式 '<bbox>x1 y1 x2 y2</bbox>'
action_inputs[param_name.strip()] = param
if "start_box" in param_name or "end_box" in param_name:
ori_box = param
# Remove parentheses and split the string by commas
numbers = ori_box.replace("(", "").replace(")", "").split(",")
# Convert to float and scale by 1000
float_numbers = [float(num) / factor for num in numbers]
if len(float_numbers) == 2:
float_numbers = [float_numbers[0], float_numbers[1], float_numbers[0], float_numbers[1]]
action_inputs[param_name.strip()] = str(float_numbers)
# import pdb; pdb.set_trace()
actions.append({
"reflection": reflection,
"thought": thought,
"action_type": action_type,
"action_inputs": action_inputs,
"text": text
})
return actions
def parsing_response_to_pyautogui_code(responses, image_height: int, image_width:int, input_swap:bool=True) -> str:
'''
将M模型的输出解析为OSWorld中的action,生成pyautogui代码字符串
参数:
response: 包含模型输出的字典,结构类似于:
{
"action_type": "hotkey",
"action_inputs": {
"hotkey": "v ctrl",
"start_box": None,
"end_box": None
}
}
返回:
生成的pyautogui代码字符串
'''
pyautogui_code = f"import pyautogui\nimport time\n"
if isinstance(responses, dict):
responses = [responses]
for response_id, response in enumerate(responses):
if "observation" in response:
observation = response["observation"]
else:
observation = ""
if "thought" in response:
thought = response["thought"]
else:
thought = ""
if response_id == 0:
pyautogui_code += f"'''\nObservation:\n{observation}\n\nThought:\n{thought}\n'''\n"
else:
pyautogui_code += f"\ntime.sleep(3)\n"
action_dict = response
action_type = action_dict.get("action_type")
action_inputs = action_dict.get("action_inputs", {})
if action_type == "hotkey":
# Parsing hotkey action
if "key" in action_inputs:
hotkey = action_inputs.get("key", "")
else:
hotkey = action_inputs.get("hotkey", "")
if hotkey:
# Handle other hotkeys
keys = hotkey.split() # Split the keys by space
pyautogui_code += f"\npyautogui.hotkey({', '.join([repr(k) for k in keys])})"
elif action_type == "type":
# Parsing typing action using clipboard
content = action_inputs.get("content", "")
content = escape_single_quotes(content)
if content:
if input_swap:
pyautogui_code += f"\nimport pyperclip"
pyautogui_code += f"\npyperclip.copy('{content.strip()}')"
pyautogui_code += f"\npyautogui.hotkey('ctrl', 'v')"
pyautogui_code += f"\ntime.sleep(0.5)\n"
if content.endswith("\n") or content.endswith("\\n"):
pyautogui_code += f"\npyautogui.press('enter')"
else:
pyautogui_code += f"\npyautogui.write('{content.strip()}', interval=0.1)"
pyautogui_code += f"\ntime.sleep(0.5)\n"
if content.endswith("\n") or content.endswith("\\n"):
pyautogui_code += f"\npyautogui.press('enter')"
elif action_type in ["drag", "select"]:
# Parsing drag or select action based on start and end_boxes
start_box = action_inputs.get("start_box")
end_box = action_inputs.get("end_box")
if start_box and end_box:
x1, y1, x2, y2 = eval(start_box) # Assuming box is in [x1, y1, x2, y2]
sx = round(float((x1 + x2) / 2) * image_width, 3)
sy = round(float((y1 + y2) / 2) * image_height, 3)
x1, y1, x2, y2 = eval(end_box) # Assuming box is in [x1, y1, x2, y2]
ex = round(float((x1 + x2) / 2) * image_width, 3)
ey = round(float((y1 + y2) / 2) * image_height, 3)
pyautogui_code += (
f"\npyautogui.moveTo({sx}, {sy})\n"
f"\npyautogui.dragTo({ex}, {ey}, duration=1.0)\n"
)
elif action_type == "scroll":
# Parsing scroll action
start_box = action_inputs.get("start_box")
if start_box:
x1, y1, x2, y2 = eval(start_box) # Assuming box is in [x1, y1, x2, y2]
x = round(float((x1 + x2) / 2) * image_width, 3)
y = round(float((y1 + y2) / 2) * image_height, 3)
# # 先点对应区域,再滚动
# pyautogui_code += f"\npyautogui.click({x}, {y}, button='left')"
else:
x = None
y = None
direction = action_inputs.get("direction", "")
if x == None:
if "up" in direction.lower():
pyautogui_code += f"\npyautogui.scroll(5)"
elif "down" in direction.lower():
pyautogui_code += f"\npyautogui.scroll(-5)"
else:
if "up" in direction.lower():
pyautogui_code += f"\npyautogui.scroll(5, x={x}, y={y})"
elif "down" in direction.lower():
pyautogui_code += f"\npyautogui.scroll(-5, x={x}, y={y})"
elif action_type in ["click", "left_single", "left_double", "right_single", "hover"]:
# Parsing mouse click actions
start_box = action_inputs.get("start_box")
start_box = str(start_box)
if start_box:
start_box = eval(start_box)
if len(start_box) == 4:
x1, y1, x2, y2 = start_box # Assuming box is in [x1, y1, x2, y2]
elif len(start_box) == 2:
x1, y1 = start_box
x2 = x1
y2 = y1
x = round(float((x1 + x2) / 2) * image_width, 3)
y = round(float((y1 + y2) / 2) * image_height, 3)
if action_type == "left_single" or action_type == "click":
pyautogui_code += f"\npyautogui.click({x}, {y}, button='left')"
elif action_type == "left_double":
pyautogui_code += f"\npyautogui.doubleClick({x}, {y}, button='left')"
elif action_type == "right_single":
pyautogui_code += f"\npyautogui.click({x}, {y}, button='right')"
elif action_type == "hover":
pyautogui_code += f"\npyautogui.moveTo({x}, {y})"
elif action_type in ["finished"]:
pyautogui_code = f"DONE"
else:
pyautogui_code += f"\n# Unrecognized action type: {action_type}"
return pyautogui_code
def pil_to_base64(image):
buffer = BytesIO()
image.save(buffer, format="PNG") # 你可以改成 "JPEG" 等格式
return base64.b64encode(buffer.getvalue()).decode("utf-8")
def linearize_accessibility_tree(accessibility_tree, platform="ubuntu"):
if platform == "ubuntu":
_attributes_ns = attributes_ns_ubuntu
_state_ns = state_ns_ubuntu
_component_ns = component_ns_ubuntu
_value_ns = value_ns_ubuntu
elif platform == "windows":
_attributes_ns = attributes_ns_windows
_state_ns = state_ns_windows
_component_ns = component_ns_windows
_value_ns = value_ns_windows
else:
raise ValueError("Invalid platform, must be 'ubuntu' or 'windows'")
filtered_nodes = filter_nodes(ET.fromstring(accessibility_tree), platform)
linearized_accessibility_tree = [
"tag\tname\ttext\tclass\tdescription\tposition (top-left x&y)\tsize (w&h)"
]
# Linearize the accessibility tree nodes into a table format
for node in filtered_nodes:
if node.text:
text = (
node.text
if '"' not in node.text
else '"{:}"'.format(node.text.replace('"', '""'))
)
elif node.get("{{{:}}}class".format(class_ns_windows), "").endswith(
"EditWrapper"
) and node.get("{{{:}}}value".format(_value_ns)):
node_text = node.get("{{{:}}}value".format(_value_ns), "")
text = (
node_text
if '"' not in node_text
else '"{:}"'.format(node_text.replace('"', '""'))
)
else:
text = '""'
linearized_accessibility_tree.append(
"{:}\t{:}\t{:}\t{:}\t{:}\t{:}\t{:}".format(
node.tag,
node.get("name", ""),
text,
(
node.get("{{{:}}}class".format(_attributes_ns), "")
if platform == "ubuntu"
else node.get("{{{:}}}class".format(class_ns_windows), "")
),
node.get("{{{:}}}description".format(_attributes_ns), ""),
node.get("{{{:}}}screencoord".format(_component_ns), ""),
node.get("{{{:}}}size".format(_component_ns), ""),
)
)
return "\n".join(linearized_accessibility_tree)
def trim_accessibility_tree(linearized_accessibility_tree, max_tokens):
# enc = tiktoken.encoding_for_model("gpt-4")
# tokens = enc.encode(linearized_accessibility_tree)
# if len(tokens) > max_tokens:
# linearized_accessibility_tree = enc.decode(tokens[:max_tokens])
# linearized_accessibility_tree += "[...]\n"
return linearized_accessibility_tree
class UITARSAgent:
def __init__(
self,
model: str,
platform="ubuntu",
max_tokens=1000,
top_p=0.9,
top_k=1.0,
temperature=0.0,
action_space="pyautogui",
observation_type="screenshot_a11y_tree",
# observation_type can be in ["screenshot", "a11y_tree", "screenshot_a11y_tree", "som"]
max_trajectory_length=50,
a11y_tree_max_tokens=10000,
runtime_conf: dict = {
"infer_mode": "qwen2vl_user",
"prompt_style": "qwen2vl_user",
"input_swap": True,
"language": "Chinese",
"max_steps": 50,
"history_n": 5,
"screen_height": 1080,
"screen_width": 1920
}
):
self.model = model
self.platform = platform
self.max_tokens = max_tokens
self.top_p = top_p
self.top_k = top_k
self.temperature = temperature
self.action_space = action_space
self.observation_type = observation_type
self.max_trajectory_length = max_trajectory_length
self.a11y_tree_max_tokens = a11y_tree_max_tokens
self.runtime_conf = runtime_conf
self.vlm = OpenAI(
base_url=os.environ['DOUBAO_API_URL'],
api_key=os.environ['DOUBAO_API_KEY'],
) # should replace with your UI-TARS server api
self.infer_mode = self.runtime_conf["infer_mode"]
self.prompt_style = self.runtime_conf["prompt_style"]
self.input_swap = self.runtime_conf["input_swap"]
self.language = self.runtime_conf["language"]
self.max_steps = max_trajectory_length
self.thoughts = []
self.actions = []
self.observations = []
self.history_images = []
self.history_responses = []
self.prompt_action_space = UITARS_ACTION_SPACE
self.customize_action_parser = parse_action_qwen2vl
self.action_parse_res_factor = 1000
if self.infer_mode == "qwen2vl_user":
self.prompt_action_space = UITARS_CALL_USR_ACTION_SPACE
self.prompt_template = UITARS_USR_PROMPT_THOUGHT
if self.prompt_style == "qwen2vl_user":
self.prompt_template = UITARS_USR_PROMPT_THOUGHT
elif self.prompt_style == "qwen2vl_no_thought":
self.prompt_template = UITARS_USR_PROMPT_NOTHOUGHT
if "history_n" in self.runtime_conf:
self.history_n = self.runtime_conf["history_n"]
else:
self.history_n = 5
def predict(
self, instruction: str, obs: Dict, last_action_after_obs: Dict = None
) -> List:
"""
Predict the next action(s) based on the current observation.
"""
# Append trajectory
# print(len(self.observations), len(self.actions), len(self.actions))
assert len(self.observations) == len(self.actions) and len(self.actions) == len(
self.thoughts
), "The number of observations and actions should be the same."
if len(self.observations) > self.max_trajectory_length:
if self.max_trajectory_length == 0:
_observations = []
_actions = []
_thoughts = []
else:
_observations = self.observations[-self.max_trajectory_length :]
_actions = self.actions[-self.max_trajectory_length :]
_thoughts = self.thoughts[-self.max_trajectory_length :]
else:
_observations = self.observations
_actions = self.actions
_thoughts = self.thoughts
if last_action_after_obs is not None and self.infer_mode == "double_image":
self.history_images.append(last_action_after_obs["screenshot"])
self.history_images.append(obs["screenshot"])
if self.observation_type in ["screenshot", "screenshot_a11y_tree"]:
base64_image = obs["screenshot"]
try:
linearized_accessibility_tree = (
linearize_accessibility_tree(
accessibility_tree=obs["accessibility_tree"],
platform=self.platform,
)
if self.observation_type == "screenshot_a11y_tree"
else None
)
except:
linearized_accessibility_tree = None
# logger.debug("LINEAR AT: %s", linearized_accessibility_tree)
if linearized_accessibility_tree:
linearized_accessibility_tree = trim_accessibility_tree(
linearized_accessibility_tree, self.a11y_tree_max_tokens
)
if self.observation_type == "screenshot_a11y_tree":
self.observations.append(
{
"screenshot": base64_image,
"accessibility_tree": linearized_accessibility_tree,
}
)
else:
self.observations.append(
{"screenshot": base64_image, "accessibility_tree": None}
)
else:
raise ValueError(
"Invalid observation_type type: " + self.observation_type
) # 1}}}
if self.infer_mode == "qwen2vl_user":
user_prompt = self.prompt_template.format(
instruction=instruction,
action_space=self.prompt_action_space,
language=self.language
)
elif self.infer_mode == "qwen2vl_no_thought":
user_prompt = self.prompt_template.format(
instruction=instruction
)
if len(self.history_images) > self.history_n:
self.history_images = self.history_images[-self.history_n:]
max_pixels = 2116800
min_pixels = 3136
messages, images = [], []
if isinstance(self.history_images, bytes):
self.history_images = [self.history_images]
elif isinstance(self.history_images, np.ndarray):
self.history_images = list(self.history_images)
elif isinstance(self.history_images, list):
pass
else:
raise TypeError(f"Unidentified images type: {type(self.history_images)}")
max_image_nums_under_32k = int(32768*0.75/max_pixels*28*28)
if len(self.history_images) > max_image_nums_under_32k:
num_of_images = min(5, len(self.history_images))
max_pixels = int(32768*0.75) // num_of_images
for turn, image in enumerate(self.history_images):
if len(images) >= 5:
break
try:
image = Image.open(BytesIO(image))
except Exception as e:
raise RuntimeError(f"Error opening image: {e}")
if image.width * image.height > max_pixels:
"""
如果图片超过/低于像素限制,则计算一个缩放因子resize_factor,使图片的像素数缩小到等于或小于max_pixels。这个缩放因子是通过开平方根计算的,确保纵横比保持不变,这样原始的相对坐标可以不经转换直接复用
"""
resize_factor = math.sqrt(max_pixels / (image.width * image.height))
width, height = int(image.width * resize_factor), int(image.height * resize_factor)
image = image.resize((width, height))
if image.width * image.height < min_pixels:
resize_factor = math.sqrt(min_pixels / (image.width * image.height))
width, height = math.ceil(image.width * resize_factor), math.ceil(image.height * resize_factor)
image = image.resize((width, height))
if image.mode != "RGB":
image = image.convert("RGB")
images.append(image)
messages = [
{
"role": "system",
"content": [{"type": "text", "text": "You are a helpful assistant."}]
},
{
"role": "user",
"content": [{"type": "text", "text": user_prompt}]
}
]
image_num = 0
if len(self.history_responses) > 0:
for history_idx, history_response in enumerate(self.history_responses):
# send at most history_n images to the model
if history_idx + self.history_n > len(self.history_responses):
cur_image = images[image_num]
encoded_string = pil_to_base64(cur_image)
messages.append({
"role": "user",
"content": [{"type": "image_url", "image_url": {"url": f"data:image/png;base64,{encoded_string}"}}]
})
image_num += 1
messages.append({
"role": "assistant",
"content": history_response
})
cur_image = images[image_num]
encoded_string = pil_to_base64(cur_image)
messages.append({
"role": "user",
"content": [{"type": "image_url", "image_url": {"url": f"data:image/png;base64,{encoded_string}"}}]
})
image_num += 1
else:
cur_image = images[image_num]
encoded_string = pil_to_base64(cur_image)
messages.append({
"role": "user",
"content": [{"type": "image_url", "image_url": {"url": f"data:image/png;base64,{encoded_string}"}}]
})
image_num += 1
try_times = 3
while True:
if try_times <= 0:
print(f"Reach max retry times to fetch response from client, as error flag.")
return "client error", ["DONE"]
try:
response = self.vlm.chat.completions.create(
model=self.model,
messages=messages,
frequency_penalty=1,
max_tokens=self.max_tokens,
temperature=self.temperature,
top_p=self.top_p
)
print("Response:")
print(response.choices[0].message.content)
prediction = response.choices[0].message.content
parsed_responses = self.customize_action_parser(
prediction,
self.action_parse_res_factor,
self.runtime_conf["screen_height"],
self.runtime_conf["screen_width"]
)
break
except Exception as e:
logger.exception(f"Error when fetching response from client, with response: {e}")
prediction = None
try_times -= 1
if prediction is None:
return "client error", ["DONE"]
self.history_responses.append(prediction)
self.thoughts.append(prediction)
try:
parsed_responses = self.customize_action_parser(
prediction,
self.action_parse_res_factor,
self.runtime_conf["screen_height"],
self.runtime_conf["screen_width"]
)
except Exception as e:
print(f"Parsing action error: {prediction}, with error:\n{e}")
return f"Parsing action error: {prediction}, with error:\n{e}", ["DONE"]
actions = []
for parsed_response in parsed_responses:
if "action_type" in parsed_response:
if parsed_response["action_type"] == FINISH_WORD:
self.actions.append(actions)
return prediction, ["DONE"]
elif parsed_response["action_type"] == WAIT_WORD:
self.actions.append(actions)
return prediction, ["WAIT"]
elif parsed_response["action_type"] == ENV_FAIL_WORD:
self.actions.append(actions)
return prediction, ["FAIL"]
elif parsed_response["action_type"] == CALL_USER:
self.actions.append(actions)
return prediction, ["FAIL"]
pyautogui_code = parsing_response_to_pyautogui_code(
parsed_response,
self.runtime_conf["screen_height"],
self.runtime_conf["screen_width"],
self.input_swap
)
actions.append(pyautogui_code)
self.actions.append(actions)
if len(self.history_responses) >= self.max_trajectory_length:
# Default to FAIL if exceed max steps
actions = ["FAIL"]
return prediction, actions
@backoff.on_exception(
backoff.constant,
# here you should add more model exceptions as you want,
# but you are forbidden to add "Exception", that is, a common type of exception
# because we want to catch this kind of Exception in the outside to ensure each example won't exceed the time limit
(
# General exceptions
SSLError,
# OpenAI exceptions
openai.RateLimitError,
openai.BadRequestError,
openai.InternalServerError,
# Google exceptions
InvalidArgument,
ResourceExhausted,
InternalServerError,
BadRequest,
# Groq exceptions
# todo: check
),
interval=30,
max_tries=10,
)
def reset(self, runtime_logger):
self.thoughts = []
self.actions = []
self.observations = []
self.history_images = []
self.history_responses = []