Selected work

Representative Papers

UAMTERS: Uncertainty-Aware Mutation Analysis for DL-enabled Robotic Software

Feb 23, 2026

Existing mutation analysis approaches struggle to effectively evaluate the testing capability of deep learning–driven robotic software under environmental uncertainty. To address this limitation, this work introduces uncertainty modeling into mutation analysis for the first time, proposing uncertainty-aware mutation operators tailored for deep learning–based robotic systems. These operators inject controllable stochastic uncertainty to simulate realistic behavioral deviations and are accompanied by novel mutation scoring metrics that quantify a test suite’s ability to detect failures across varying levels of uncertainty. Experimental evaluation on three robotic case studies demonstrates that the proposed method more accurately discriminates between test suite qualities and effectively captures software failures induced by environmental uncertainty.

0 citationsRead paper

HEXAR: a Hierarchical Explainability Architecture for Robots

Jan 06, 2026arXiv.org

This work addresses the challenge of balancing high-level behavioral understanding with modular architecture in robot explainability, a trade-off that often confines existing methods to either overly localized or excessively holistic explanations. To overcome this limitation, the authors propose a plug-and-play hierarchical explainability framework that equips individual robot modules with dedicated explainers and employs an explainer selector to dynamically orchestrate query-specific optimal explanations. For the first time, this framework integrates modularity, hierarchy, and diverse explanation techniques—including large language model (LLM) reasoning, causal models, and feature importance—enabling precise, efficient, and customizable interpretations of complex robotic behaviors. Evaluated on TIAGo assistive tasks for elderly care, the approach significantly outperforms end-to-end and aggregated baselines in root-cause identification, erroneous information filtering, and computational efficiency.

0 citationsRead paper

Out of Distribution Detection in Self-adaptive Robots with AI-powered Digital Twins

Sep 16, 2025

Addressing the challenge of proactive out-of-distribution (OOD) anomaly detection for adaptive robots operating in complex, uncertain environments, this paper proposes ODiSAR—an AI-driven digital twin framework for OOD detection. ODiSAR integrates a Transformer-based digital twin model, Monte Carlo Dropout, and reconstruction error analysis to jointly quantify both prediction error and epistemic uncertainty. Crucially, it introduces an explainability layer that maps anomalous decisions to specific state dimensions, enabling interpretable, autonomous robot adaptation. Evaluated on industrial robotic manipulation and maritime navigation tasks, ODiSAR achieves 98% AUROC, 96% true negative rate at 95% true positive rate (TNR@TPR95), and 95% F1-score—significantly improving both detection accuracy and interpretability for previously unseen anomalies. This work establishes a novel paradigm for robust operation of situated adaptive robots (SAR) under distributional shift.

0 citationsRead paper

Quantum Machine Learning-based Test Oracle for Autonomous Mobile Robots

Aug 04, 2025

Autonomous mobile robots operating in unknown dynamic environments lack reliable test oracles—i.e., accurate specifications of expected correct behavior—hindering effective verification and regression testing. Method: This paper proposes QuReBot, a quantum-classical hybrid framework that uniquely integrates quantum reservoir computing with a classical neural architecture inspired by residual networks, thereby overcoming the non-convergence issues prevalent in purely quantum models. QuReBot jointly models environmental perception and motion behavior to enhance oracle generation. Contribution/Results: Experimental evaluation demonstrates that QuReBot reduces prediction error by 15% over classical baseline models. Its generalizability and tunability are validated across multiple robot configurations. Moreover, the framework provides practical, deployment-oriented guidelines for parameter optimization, enabling robust real-world application.

0 citationsRead paper

Vision Language Model-based Testing of Industrial Autonomous Mobile Robots

Aug 04, 2025

Industrial autonomous mobile robots (AMRs) face high-cost, high-risk, and low-coverage safety testing in human-shared environments. To address this, we propose RVSG—a novel method that introduces vision-language models (VLMs) into AMR testing for the first time, enabling automated synthesis of semantically guided, safety-violating human behavioral scenarios from natural-language functional and safety requirements. RVSG leverages VLMs to interpret requirements, generate diverse human–robot interaction perturbations, and validate them in closed-loop simulation with a PAL Robotics AMR. Experimental results demonstrate that RVSG significantly improves both efficiency and diversity in generating requirement-violating scenarios: robot behavioral variability increases by 37%, and defect detection rate rises by 2.1× compared to baseline methods. Crucially, RVSG effectively uncovers previously unknown safety vulnerabilities arising from unanticipated human–robot interactions.

0 citationsRead paper

Recent publications

Latest Papers

UAMTERS: Uncertainty-Aware Mutation Analysis for DL-enabled Robotic Software

Feb 23, 2026

Existing mutation analysis approaches struggle to effectively evaluate the testing capability of deep learning–driven robotic software under environmental uncertainty. To address this limitation, this work introduces uncertainty modeling into mutation analysis for the first time, proposing uncertainty-aware mutation operators tailored for deep learning–based robotic systems. These operators inject controllable stochastic uncertainty to simulate realistic behavioral deviations and are accompanied by novel mutation scoring metrics that quantify a test suite’s ability to detect failures across varying levels of uncertainty. Experimental evaluation on three robotic case studies demonstrates that the proposed method more accurately discriminates between test suite qualities and effectively captures software failures induced by environmental uncertainty.

0 citationsRead paper

HEXAR: a Hierarchical Explainability Architecture for Robots

Jan 06, 2026arXiv.org

This work addresses the challenge of balancing high-level behavioral understanding with modular architecture in robot explainability, a trade-off that often confines existing methods to either overly localized or excessively holistic explanations. To overcome this limitation, the authors propose a plug-and-play hierarchical explainability framework that equips individual robot modules with dedicated explainers and employs an explainer selector to dynamically orchestrate query-specific optimal explanations. For the first time, this framework integrates modularity, hierarchy, and diverse explanation techniques—including large language model (LLM) reasoning, causal models, and feature importance—enabling precise, efficient, and customizable interpretations of complex robotic behaviors. Evaluated on TIAGo assistive tasks for elderly care, the approach significantly outperforms end-to-end and aggregated baselines in root-cause identification, erroneous information filtering, and computational efficiency.

0 citationsRead paper

Out of Distribution Detection in Self-adaptive Robots with AI-powered Digital Twins

Sep 16, 2025

Addressing the challenge of proactive out-of-distribution (OOD) anomaly detection for adaptive robots operating in complex, uncertain environments, this paper proposes ODiSAR—an AI-driven digital twin framework for OOD detection. ODiSAR integrates a Transformer-based digital twin model, Monte Carlo Dropout, and reconstruction error analysis to jointly quantify both prediction error and epistemic uncertainty. Crucially, it introduces an explainability layer that maps anomalous decisions to specific state dimensions, enabling interpretable, autonomous robot adaptation. Evaluated on industrial robotic manipulation and maritime navigation tasks, ODiSAR achieves 98% AUROC, 96% true negative rate at 95% true positive rate (TNR@TPR95), and 95% F1-score—significantly improving both detection accuracy and interpretability for previously unseen anomalies. This work establishes a novel paradigm for robust operation of situated adaptive robots (SAR) under distributional shift.

0 citationsRead paper

Quantum Machine Learning-based Test Oracle for Autonomous Mobile Robots

Aug 04, 2025

Autonomous mobile robots operating in unknown dynamic environments lack reliable test oracles—i.e., accurate specifications of expected correct behavior—hindering effective verification and regression testing. Method: This paper proposes QuReBot, a quantum-classical hybrid framework that uniquely integrates quantum reservoir computing with a classical neural architecture inspired by residual networks, thereby overcoming the non-convergence issues prevalent in purely quantum models. QuReBot jointly models environmental perception and motion behavior to enhance oracle generation. Contribution/Results: Experimental evaluation demonstrates that QuReBot reduces prediction error by 15% over classical baseline models. Its generalizability and tunability are validated across multiple robot configurations. Moreover, the framework provides practical, deployment-oriented guidelines for parameter optimization, enabling robust real-world application.

0 citationsRead paper

Vision Language Model-based Testing of Industrial Autonomous Mobile Robots

Aug 04, 2025

Industrial autonomous mobile robots (AMRs) face high-cost, high-risk, and low-coverage safety testing in human-shared environments. To address this, we propose RVSG—a novel method that introduces vision-language models (VLMs) into AMR testing for the first time, enabling automated synthesis of semantically guided, safety-violating human behavioral scenarios from natural-language functional and safety requirements. RVSG leverages VLMs to interpret requirements, generate diverse human–robot interaction perturbations, and validate them in closed-loop simulation with a PAL Robotics AMR. Experimental results demonstrate that RVSG significantly improves both efficiency and diversity in generating requirement-violating scenarios: robot behavioral variability increases by 37%, and defect detection rate rises by 2.1× compared to baseline methods. Crucially, RVSG effectively uncovers previously unknown safety vulnerabilities arising from unanticipated human–robot interactions.

0 citationsRead paper