Reasoning and Planning for Long-term Active Embodied Question Answering

View a PDF file from the paper entitled Enter the Palace mind: Thinking and planning to answer the long -term questions, by Muhammad Fadihil and 12 other books

PDF HTML (experimental) view

a summary:When robots are increasingly able to work for long periods – spanning days, weeks and even months – knowledge is expected to accumulate with their environments and benefit from this experience to help humans more effectively. This paper studies the problem of answering the long-term active questions (La-eqa), a new task in which the robot must remember previous experiences and explore its environment actively to answer temporarily and temporarily misleading questions. Unlike the traditional EQA settings, which usually focus on understanding the current environment alone or on calling the previous observation, La-eqa challenges an agent to cause cases of past, present and possible future, setting time for exploration, the date for consulting its memory, and when it stops collecting notes and providing a final answer. Standard EQA approaches are based on the large models in this preparation due to limited windows, the absence of continuous memory, and the inability to combine memory summons with active exploration. To address this, we suggest a system of systematic memory for robots, inspired by the way the mind is short of cognitive sciences. Our method encodes occasional experiences as global scenery, and constitutes the thinking and planning algorithm that allows the recovery of the targeted memory and guided navigation. To balance the re -exploration comparison, we provide the criteria for stopping the value -based value that determines when the agent collects sufficient information. We evaluate our way of real world experiences and offer a new standard that extends popular simulation environments and actual industrial sites. Our approach is greatly outperforming modern foundation lines, which leads to great gains in both the accuracy of the answer and the efficiency of exploration.