Selected work

Representative Papers

Defending Retrieval-Augmented Intrusion Detection Against Knowledge Poisoning and Prompt Injection

Aug 08, 2026

This work addresses the vulnerability of existing retrieval-augmented intrusion detection systems (RAG-IDS) to knowledge poisoning and prompt injection attacks, which degrade performance and induce high false-positive rates. To mitigate these threats, the authors propose a novel three-tier multi-agent RAG-IDS framework that integrates soft trust scoring, label embedding consistency checking (LECC), and prompt sanitization mechanisms to establish a robust retrieval-boundary defense. Experimental results demonstrate substantial improvements in system robustness: under 30% knowledge poisoning on the CIC-UNSW-NB15 dataset, the approach achieves a classification performance recovery rate of 0.57; under prompt injection attacks, the label-flipping success rate is reduced to 0.6–2.4%, significantly outperforming single-document retrieval baselines, which exhibit rates of 35–55%.

0 citationsRead paper

SciRet: A Compute-Aware Empirical Study of Retrieval and Reranking for Scientific RAG

Aug 04, 2026

This study systematically evaluates the effectiveness and efficiency of hybrid retrieval and re-ranking strategies in Retrieval-Augmented Generation (RAG) across varying scales of scientific corpora—specifically 1K, 5K, and 15K CORD-19 papers. The approach integrates sentence-window chunking, BM25 and BGE-M3 dense retrieval, reciprocal rank fusion, optional cross-encoder re-ranking, and evidence-based answer generation. Results demonstrate that hybrid retrieval achieves perfect Recall@10 (1.000) at both the 1K and 15K corpus scales, while RAGAS faithfulness improves with larger corpus sizes. Notably, cross-encoders trained on general-domain data degrade accuracy due to domain mismatch. The work introduces a controllable evaluation paradigm based on pseudo relevance labels, revealing the superior performance of hybrid retrieval for scientific question answering.

0 citationsRead paper

Simple but not Simpler: A Surface-Sliding Method for Finding the Minimum Distance between Two Ellipsoids

Mar 23, 2026

This work addresses the problem of efficiently and accurately computing the minimum distance between two ellipsoids. The authors propose an iterative algorithm based on surface sliding, which operates in the θ–φ parametric space by updating the positions of two points—one on each ellipsoid—guided by the geometric tension of the connecting line segment. The iteration proceeds until this segment becomes simultaneously normal to both ellipsoidal surfaces, ensuring convergence to the true minimal distance. The method features a clear geometric interpretation, concise analytical expressions, and low computational complexity, and it naturally extends to other smooth convex bodies. Experimental results demonstrate that the proposed approach outperforms existing methods in terms of accuracy, stability, consistency, and robustness.

0 citationsRead paper

Trust in Autonomous Human--Robot Collaboration: Effects of Responsive Interaction Policies

Feb 25, 2026

This study investigates how interaction strategies influence the formation of human trust in fully autonomous human-robot collaboration. Conducted within a genuinely autonomous system—without Wizard-of-Oz intervention—the research compares responsive and neutral-reactive strategies, integrating autonomous spoken dialogue management, real-time affect inference, and adaptive interaction mechanisms to dynamically assess trust across multi-stage collaborative tasks. The findings reveal that trust is jointly constrained by the dynamics of interaction and communication effectiveness: the responsive strategy significantly enhances trust when communication is efficient, yet this advantage diminishes as linguistic interaction quality degrades. Moreover, affective and experiential dimensions of trust are more vulnerable to communication disruptions than reliability. This work provides the first empirical elucidation, under fully autonomous conditions, of the dynamic relationship between social robot interaction strategies and human trust.

0 citationsRead paper

A Modular Object Detection System for Humanoid Robots Using YOLO

Oct 15, 2025

To address the low efficiency and poor robustness of visual detection in humanoid robots under computationally constrained conditions, this paper proposes a modular object detection system based on YOLOv9. The system is deeply integrated into the ROS 1 framework and specifically designed for the dynamic competition environment of FIRA HuroCup. It introduces a lightweight, extensible, virtualized vision module that enables efficient simulation-to-real transfer. Through custom dataset training and inference optimization, the system achieves mean Average Precision (mAP) comparable to conventional geometric methods while significantly improving detection robustness and real-time performance—evidenced by substantial FPS gains—in complex, dynamic scenarios. Experimental results demonstrate a favorable trade-off among accuracy, speed, and environmental adaptability. The proposed solution provides a reusable, engineering-ready embedded vision framework for humanoid robots.

0 citationsRead paper

Recent publications

Latest Papers

Defending Retrieval-Augmented Intrusion Detection Against Knowledge Poisoning and Prompt Injection

Aug 08, 2026

This work addresses the vulnerability of existing retrieval-augmented intrusion detection systems (RAG-IDS) to knowledge poisoning and prompt injection attacks, which degrade performance and induce high false-positive rates. To mitigate these threats, the authors propose a novel three-tier multi-agent RAG-IDS framework that integrates soft trust scoring, label embedding consistency checking (LECC), and prompt sanitization mechanisms to establish a robust retrieval-boundary defense. Experimental results demonstrate substantial improvements in system robustness: under 30% knowledge poisoning on the CIC-UNSW-NB15 dataset, the approach achieves a classification performance recovery rate of 0.57; under prompt injection attacks, the label-flipping success rate is reduced to 0.6–2.4%, significantly outperforming single-document retrieval baselines, which exhibit rates of 35–55%.

0 citationsRead paper

SciRet: A Compute-Aware Empirical Study of Retrieval and Reranking for Scientific RAG

Aug 04, 2026

This study systematically evaluates the effectiveness and efficiency of hybrid retrieval and re-ranking strategies in Retrieval-Augmented Generation (RAG) across varying scales of scientific corpora—specifically 1K, 5K, and 15K CORD-19 papers. The approach integrates sentence-window chunking, BM25 and BGE-M3 dense retrieval, reciprocal rank fusion, optional cross-encoder re-ranking, and evidence-based answer generation. Results demonstrate that hybrid retrieval achieves perfect Recall@10 (1.000) at both the 1K and 15K corpus scales, while RAGAS faithfulness improves with larger corpus sizes. Notably, cross-encoders trained on general-domain data degrade accuracy due to domain mismatch. The work introduces a controllable evaluation paradigm based on pseudo relevance labels, revealing the superior performance of hybrid retrieval for scientific question answering.

0 citationsRead paper

Simple but not Simpler: A Surface-Sliding Method for Finding the Minimum Distance between Two Ellipsoids

Mar 23, 2026

This work addresses the problem of efficiently and accurately computing the minimum distance between two ellipsoids. The authors propose an iterative algorithm based on surface sliding, which operates in the θ–φ parametric space by updating the positions of two points—one on each ellipsoid—guided by the geometric tension of the connecting line segment. The iteration proceeds until this segment becomes simultaneously normal to both ellipsoidal surfaces, ensuring convergence to the true minimal distance. The method features a clear geometric interpretation, concise analytical expressions, and low computational complexity, and it naturally extends to other smooth convex bodies. Experimental results demonstrate that the proposed approach outperforms existing methods in terms of accuracy, stability, consistency, and robustness.

0 citationsRead paper

Trust in Autonomous Human--Robot Collaboration: Effects of Responsive Interaction Policies

Feb 25, 2026

This study investigates how interaction strategies influence the formation of human trust in fully autonomous human-robot collaboration. Conducted within a genuinely autonomous system—without Wizard-of-Oz intervention—the research compares responsive and neutral-reactive strategies, integrating autonomous spoken dialogue management, real-time affect inference, and adaptive interaction mechanisms to dynamically assess trust across multi-stage collaborative tasks. The findings reveal that trust is jointly constrained by the dynamics of interaction and communication effectiveness: the responsive strategy significantly enhances trust when communication is efficient, yet this advantage diminishes as linguistic interaction quality degrades. Moreover, affective and experiential dimensions of trust are more vulnerable to communication disruptions than reliability. This work provides the first empirical elucidation, under fully autonomous conditions, of the dynamic relationship between social robot interaction strategies and human trust.

0 citationsRead paper

A Modular Object Detection System for Humanoid Robots Using YOLO

Oct 15, 2025

To address the low efficiency and poor robustness of visual detection in humanoid robots under computationally constrained conditions, this paper proposes a modular object detection system based on YOLOv9. The system is deeply integrated into the ROS 1 framework and specifically designed for the dynamic competition environment of FIRA HuroCup. It introduces a lightweight, extensible, virtualized vision module that enables efficient simulation-to-real transfer. Through custom dataset training and inference optimization, the system achieves mean Average Precision (mAP) comparable to conventional geometric methods while significantly improving detection robustness and real-time performance—evidenced by substantial FPS gains—in complex, dynamic scenarios. Experimental results demonstrate a favorable trade-off among accuracy, speed, and environmental adaptability. The proposed solution provides a reusable, engineering-ready embedded vision framework for humanoid robots.

0 citationsRead paper