thomas-yanxin/robovqa-mirror

Dataset Structure

Data Fields

• uid (string): Unique identifier for the episode.
• video (string): Path or URL to the video file.
• task_type (string): Type of task (e.g., planning, success evaluation).
• question (string): Natural language question.
• answer (string): Ground truth answer (free-form or binary).

Preprocess

def parse_task_data(text_data: str) -> list:
    """
    Parses a string containing task data to extract task type, question, and answer.
    Handles multiple Q: A: pairs within a single text block.

    Args:
        text_data: The input string containing the task information.

    Returns:
        A list of dictionaries, where each dictionary represents a parsed task
        and contains 'task_type', 'question', and 'answer' keys.
    """
    parsed_results = []

    # Split the input into multiple <task:...> blocks
    task_blocks = re.findall(r'(<task:[^>]+>.*?)(?=<task:|$)', text_data, re.DOTALL)

    for block in task_blocks:
        # Extract task type
        task_type_match = re.search(r'<task:([^>]+)>', block)
        task_type = task_type_match.group(1) if task_type_match else "unknown"

        # Remove the task tag for easier processing
        clean_block = re.sub(r'<task:[^>]+>', '', block, 1).strip()

        # Match everything from start up to <PRED>A: as question, then capture answer
        qa_pairs = re.findall(r'(.*?)Q: (.*?) <PRED>A: (.*?)</PRED>', clean_block, re.DOTALL)

        for prefix, q_suffix, raw_answer in qa_pairs:
            # Combine both parts of the question
            question = (prefix + "Q: " + q_suffix).strip()

            # Clean answer by removing nested tags
            answer = re.sub(
                r'<PRED:ANSWER>|<PRED:DISCRETE>|<PRED:BINARY>|</PRED:BINARY>|</PRED:DISCRETE>|</PRED:ANSWER>|\n',
                '',
                raw_answer
            ).strip()

            parsed_results.append({
                "question_type": task_type,
                "question": question,
                "answer": answer
            })

    return parsed_results