Explainable Autonomous Robots: Taxonomy, State, and Trends — An Implementation-driven Mapping Study

Abstract

As autonomous robots increasingly operate in dynamic environments shared with humans, explaining their decisions and behaviors is essential for trust, transparency, and effective human-robot interaction. Although eXplainable Artificial Intelligence (XAI) has received substantial attention in recent years, challenges specific to eXplainable Autonomous Robot (XAR) remain underexplored. We report a systematic mapping study of 94 papers (922 screened) published between 2021 and February 2026, providing empirical evidence on how XAR research is implemented. Based on the analyzed papers, we derive a taxonomy spanning target platforms, software characteristics, explanation structures, stakeholders, and evaluation practices. With this, we categorize the reviewed literature and derive insights, trends and challenges for XAR research.The field demonstrates technical feasibility across diverse contexts. However, we identify clear opportunities: reversing the current trend from real-world validation to simulations (especially in Machine Learning (ML)-based systems), more explicit specification of the intended context of use, deeper investigation of human involvement, exploration of multi-agent/multi-stakeholder scenarios, and closer alignment between researchers’ focus on “Why” explanations and users’ broader question needs. Furthermore, current evaluation practices are inconsistent in their initial motivation and measured results. Finally, XAR research forgoes comparative analysis and reproducibility due to a severe lack of open-source code.