Navigating an aircraft is a complicated problem as it requires analyzing extensive operational data to identify an optimal and safe path. Recently, many studies have employed deep reinforcement learning (DRL) to address the navigation problem, and it has been demonstrated to be a powerful tool for addressing these kinds of problems. Despite being a promising tool, the opaque nature of the reasoning mechanisms of DRL methods make them difficult to troubleshoot performance issues and fosters mistrust, thereby constraining their application in critical scenarios. This deficiency necessitates the need for the development of explanation methods aimed at clarifying the reasoning processes of DRL models for users. Additionally, governments around the world are beginning to require the inclusion of such explanation methods in autonomous systems used for safety critical applications. In this paper, we propose NavChat, an explanation framework designed to justify the trajectory suggested by a trained DRL model for an efficient and safe path in an unknown environment. NavChat is a post-hoc approach that leverages SHapley Additive exPlanations (SHAP), a Large Language Model (LLM), and prompt engineering to generate natural language explanations. Using SHAP, we identify the most important features that affect the agent's decision. The LLM provides natural language explanations which facilitate faster comprehension and more lucidity for non-expert users in comparison with visual explanations. We compare the explanations provided by NavChat for various prompts and then evaluate the accuracy and consistency of its explanations via a user study. This method facilitates understanding of the decisions made by the automated navigation model and communicates the reasoning process.


    Access

    Check access

    Check availability in my library

    Order at Subito €


    Export, share and cite



    Title :

    Natural Language Explanation for Autonomous Navigation


    Contributors:


    Publication date :

    2024-09-29


    Size :

    1089738 byte





    Type of media :

    Conference paper


    Type of material :

    Electronic Resource


    Language :

    English



    INDOOR AUTONOMOUS NAVIGATION BASED ON NATURAL LANGUAGE

    ZHAO JINXIN / ZHANG LIANGJUN | European Patent Office | 2022

    Free access

    Natural feature tracking for autonomous navigation

    Paar,G. / Sidla,O. / Poelzleitner,W. et al. | Automotive engineering | 1995


    Natural Feature Tracking for Autonomous Navigation

    Paar, G. / Sidla, O. / Poelzleitner, W. | British Library Conference Proceedings | 1995


    Autonomous vehicle post-action explanation system

    European Patent Office | 2023

    Free access

    AUTONOMOUS VEHICLE POST-ACTION EXPLANATION SYSTEM

    SAHAWNEH LAITH / CSERNA BENCE | European Patent Office | 2023

    Free access