HEXAR: a Hierarchical Explainability Architecture for Robots

TL;DR

HEXAR addresses the need for explainability in modular robotic systems by introducing a Hierarchical EXplainability Architecture for Robots that coordinates multiple specialized explainers through an explainer selector. Deployed as a plug-in to existing robotic architectures, HEXAR leverages diverse explanation techniques tailored to specific modules (e.g., LLMs for planning, causal models for interaction, LIME for feature importance) and aggregates their outputs as needed. In a TIAGo home-assistant use case, HEXAR outperforms end-to-end and all-components baselines across 180 scenario-query variations, achieving higher root-cause identification accuracy and lower incidence of incorrect information with faster explanations. The results validate the modular, hierarchical approach as a scalable path toward transparent autonomous robots and underscore the importance of selective aggregation over exhaustive, monolithic explanation systems.

Abstract

As robotic systems become increasingly complex, the need for explainable decision-making becomes critical. Existing explainability approaches in robotics typically either focus on individual modules, which can be difficult to query from the perspective of high-level behaviour, or employ monolithic approaches, which do not exploit the modularity of robotic architectures. We present HEXAR (Hierarchical EXplainability Architecture for Robots), a novel framework that provides a plug-in, hierarchical approach to generate explanations about robotic systems. HEXAR consists of specialised component explainers using diverse explanation techniques (e.g., LLM-based reasoning, causal models, feature importance, etc) tailored to specific robot modules, orchestrated by an explainer selector that chooses the most appropriate one for a given query. We implement and evaluate HEXAR on a TIAGo robot performing assistive tasks in a home environment, comparing it against end-to-end and aggregated baseline approaches across 180 scenario-query variations. We observe that HEXAR significantly outperforms baselines in root cause identification, incorrect information exclusion, and runtime, offering a promising direction for transparent autonomous systems.

Figures (6)

• Figure 1: Our novel framework, HEXAR, in which different specialised component explainers are orchestrated by an explainer selector, which selects an appropriate component explainer to answer a given query.
• Figure 2: Use case of a Tiago robot that assists in a home environment.
• Figure 3: The HEXAR framework applied to the home assistant use case, showing the dependencies between modules. A component explainer is implemented for each robot module, employing a variety of explanation techniques.
• Figure 4: Evaluated baselines: end-to-end (top) and all components (bottom).
• Figure 5: Mean scores for each method, grouped by metric. For each pairwise difference, $*$ denotes $p<0.05$ and $***$ denotes $p<0.001$. Unlabelled pairs lack statistically significant difference.