Personalised Explanations in Long-term Human-Robot Interactions

Ferran Gebellí, Anaís Garrell, Jan-Gerrit Habekost, Séverin Lemaignan, Stefan Wermter, Raquel Ros

Abstract

In the field of Human-Robot Interaction (HRI), a fundamental challenge is to facilitate human understanding of robots. The emerging domain of eXplainable HRI (XHRI) investigates methods to generate explanations and evaluate their impact on human-robot interactions. Previous works have highlighted the need to personalise the level of detail of these explanations to enhance usability and comprehension. Our paper presents a framework designed to update and retrieve user knowledge-memory models, allowing for adapting the explanations' level of detail while referencing previously acquired concepts. Three architectures based on our proposed framework that use Large Language Models (LLMs) are evaluated in two distinct scenarios: a hospital patrolling robot and a kitchen assistant robot. Experimental results demonstrate that a two-stage architecture, which first generates an explanation and then personalises it, is the framework architecture that effectively reduces the level of detail only when there is related user knowledge.

Introduction

In eXplainable Human-Robot Interaction (XHRI), robots must explain their decisions to foster trust and understanding. However, current systems often provide static, repetitive explanations. In long-term scenarios—such as a robot assistant in a hospital or kitchen—this redundancy can lead to cognitive overload and user frustration. The core challenge is personalization: a robot should estimate the user's "mental model" and adjust its level of detail accordingly. If a robot has previously explained a technical limitation, it should reference that knowledge in future interactions rather than repeating the full explanation.

Conceptual overview of the personalization framework over time, showing how the robot references past knowledge to provide concise explanations.

The Personalization Framework

The proposed framework integrates four key modules to manage user knowledge:

  1. Persistent User Memory: Stores concepts (e.g., "The robot can be obstructed by obstacles") as text embeddings.
  2. Knowledge Extraction: An LLM extracts key facts from the robot's explanations to update the user's memory.
  3. Retrieval of Relevant Knowledge: Uses Retrieval-Augmented Generation (RAG) with a similarity threshold ($\tau_{retr}$) to find related concepts in memory.
  4. Post-processing (The Forgetting Curve): Applies the Ebbinghaus forgetting curve ($P = e^{-\lambda t}$) to model the probability that a user still remembers a concept based on elapsed time, ensuring the robot provides more detail if the information is likely forgotten.

Architectural Variants

The researchers compared three architectural variants for the retrieval query to determine which best bridges the gap between user questions and robot internal states:

  • Question Query: Uses the user's raw question to search memory.
  • Robot Context Query: Uses the robot's internal logs to search memory.
  • Explanation Query (Two-Stage): First generates a full explanation, then uses that explanation to search memory for overlaps, and finally adapts the text into a personalized version.
Comparison of the three proposed architectures: Question Query (top), Robot Context Query (middle), and the Two-Stage Explanation Query (bottom).

Experimental Evaluation

The framework was evaluated using synthetic data across two use cases: a kitchen assistant (NICOL robot) and a hospital patrolling robot (Tiago LITE). These scenarios involved various failure modes, such as perception errors or hardware issues like motor overheating.

Key Results

  • Superiority of the Two-Stage Approach: The "Explanation Query" architecture performed best, achieving retrieval precision between 94.27% and 96.15%. By generating the full explanation first, the system creates a semantic bridge that more effectively retrieves relevant past knowledge than the user's question alone.
  • Adaptive Length Reduction: For topics learned recently (e.g., 1 week ago), the system reduced explanation length by approximately 39%. Conversely, for new topics, the system correctly identified the lack of prior knowledge, resulting in a negligible length reduction (~2.5%) and ensuring the user received full information.
  • Robustness and Forgetting: The integration of the forgetting curve allows the robot to be "patient," automatically providing more detail if a concept hasn't been discussed for a significant period (e.g., one month).

Conclusion

This research demonstrates that a two-stage LLM architecture can effectively personalize robot explanations by leveraging persistent user memory and modeling human forgetting. While the two-stage process increases inference latency (averaging 3.88 seconds), it significantly improves the relevance and conciseness of interactions. Future work will involve longitudinal user studies to evaluate the impact of this framework on human trust and long-term usability.

Summary updated

Abstract

In the field of Human-Robot Interaction (HRI), a fundamental challenge is to facilitate human understanding of robots. The emerging domain of eXplainable HRI (XHRI) investigates methods to generate explanations and evaluate their impact on human-robot interactions. Previous works have highlighted the need to personalise the level of detail of these explanations to enhance usability and comprehension. Our paper presents a framework designed to update and retrieve user knowledge-memory models, allowing for adapting the explanations' level of detail while referencing previously acquired concepts. Three architectures based on our proposed framework that use Large Language Models (LLMs) are evaluated in two distinct scenarios: a hospital patrolling robot and a kitchen assistant robot. Experimental results demonstrate that a two-stage architecture, which first generates an explanation and then personalises it, is the framework architecture that effectively reduces the level of detail only when there is related user knowledge.
BibTeX
@inproceedings{gebelli2025personalised,
  author = {Ferran Gebellí and Anaís Garrell and Jan-Gerrit Habekost and Séverin Lemaignan and Stefan Wermter and Raquel Ros},
  title = {{Personalised Explanations in Long-term Human-Robot Interactions}},
  booktitle = {RO-MAN},
  year = {2025},
  doi = {10.1109/ro-man63969.2025.11217900},
}
Figure