Institution profile

Institut de Recherche en Communications et Cybernétique de Nantes

Academic institutioneurope · fr
Official website
Research library52linked papers
Opportunities0open roles
Selected work

Representative Papers

MSA-technique for stiffness modeling of manipulators with complex and hybrid structures

Nov 19, 2025IFAC Symposium on Robot Control

Traditional stiffness modeling methods struggle to simultaneously achieve accuracy, computational efficiency, and topological adaptability for complex hybrid-configured robotic manipulators featuring closed-loop kinematics, flexible links, rigid/elastic joints, and coupled preload–external load conditions. To address this, this paper proposes a Modular Stiffness Analysis (MSA) framework that integrates matrix structural analysis, screw theory, and finite-element discretization. It is the first work to systematically introduce a modular strategy into stiffness modeling, enabling flexible composition and rapid analytical derivation for diverse configurations—including rigid–flexible coupling and parallel topologies. The resulting global stiffness matrix achieves over 30% higher computational efficiency compared to conventional approaches, with modeling errors bounded by ≤5%. Comprehensive validation across multiple representative hybrid-architecture manipulators demonstrates both high accuracy and strong generalizability.

12 citationsRead paper

What Makes a Good Layer? Assessing the Layer-Wise Intrinsic Properties of Music Foundation Models

Aug 14, 2026

This study addresses the lack of theoretical grounding for layer selection and the ineffectiveness of existing metrics in tonal tasks involving music foundation models. By systematically analyzing inter-layer properties across twelve models, we propose pitch-shift equivariance as an unsupervised proxy metric for layer selection. This measure effectively compensates for the limitations of general evaluations in capturing tonal representations. Experimental results demonstrate that the proposed metric aligns closely with downstream task performance. Furthermore, in few-shot scenarios, it matches or outperforms trainable multi-layer fusion approaches. Consequently, this work provides a reliable theoretical basis and practical guidance for applying music foundation models to tonal tasks, bridging the gap between representation analysis and effective layer utilization without reliance on labeled data.

0 citationsRead paper
Recent publications

Latest Papers

What Makes a Good Layer? Assessing the Layer-Wise Intrinsic Properties of Music Foundation Models

Aug 14, 2026

This study addresses the lack of theoretical grounding for layer selection and the ineffectiveness of existing metrics in tonal tasks involving music foundation models. By systematically analyzing inter-layer properties across twelve models, we propose pitch-shift equivariance as an unsupervised proxy metric for layer selection. This measure effectively compensates for the limitations of general evaluations in capturing tonal representations. Experimental results demonstrate that the proposed metric aligns closely with downstream task performance. Furthermore, in few-shot scenarios, it matches or outperforms trainable multi-layer fusion approaches. Consequently, this work provides a reliable theoretical basis and practical guidance for applying music foundation models to tonal tasks, bridging the gap between representation analysis and effective layer utilization without reliance on labeled data.

0 citationsRead paper

Global Attention-Fused Image Cropping with Attention-Guided and Global-Aligned Crop Evaluator

Aug 05, 2026

Existing image cropping methods overly emphasize salient regions while neglecting the global structural relationships among primary compositional elements, leading to suboptimal aesthetic evaluation. To address this limitation, this work proposes the GAFIC framework, which introduces an Attention-Guided Feature Fusion (AGFF) mechanism to jointly model local details and global composition. Furthermore, a Global Alignment Cropping Evaluator (GACE) is designed to assess the consistency between candidate crops and the image’s overall structure, complemented by a multi-scale ranking loss to refine cropping scores. Notably, the method operates without altering pixel content, instead selecting the optimal cropping region. Extensive experiments demonstrate that GAFIC significantly outperforms state-of-the-art approaches on both the GAIC and CPC datasets, achieving superior performance in terms of accuracy, stability, and batch-processing efficiency.

0 citationsRead paper