Guardado en:
Detalles Bibliográficos
Autor principal: Lu, Yuhong
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:https://arxiv.org/abs/2509.14383
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866912591398305792
author Lu, Yuhong
author_facet Lu, Yuhong
contents Unified multi-modal encoders that bind vision, audio, and other sensors into a shared embedding space are attractive building blocks for robot perception and decision-making. However, on-robot deployment exposes the vision branch to adversarial and natural corruptions, making robustness a prerequisite for safety. Prior defenses typically align clean and adversarial features within CLIP-style encoders and overlook broader cross-modal correspondence, yielding modest gains and often degrading zero-shot transfer. We introduce RLBind, a two-stage adversarial-invariant cross-modal alignment framework for robust unified embeddings. Stage 1 performs unsupervised fine-tuning on clean-adversarial pairs to harden the visual encoder. Stage 2 leverages cross-modal correspondence by minimizing the discrepancy between clean/adversarial features and a text anchor, while enforcing class-wise distributional alignment across modalities. Extensive experiments on Image, Audio, Thermal, and Video data show that RLBind consistently outperforms the LanguageBind backbone and standard fine-tuning baselines in both clean accuracy and norm-bounded adversarial robustness. By improving resilience without sacrificing generalization, RLBind provides a practical path toward safer multi-sensor perception stacks for embodied robots in navigation, manipulation, and other autonomy settings.
format Preprint
id arxiv_https___arxiv_org_abs_2509_14383
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle RLBind: Adversarial-Invariant Cross-Modal Alignment for Unified Robust Embeddings
Lu, Yuhong
Robotics
Computer Vision and Pattern Recognition
Unified multi-modal encoders that bind vision, audio, and other sensors into a shared embedding space are attractive building blocks for robot perception and decision-making. However, on-robot deployment exposes the vision branch to adversarial and natural corruptions, making robustness a prerequisite for safety. Prior defenses typically align clean and adversarial features within CLIP-style encoders and overlook broader cross-modal correspondence, yielding modest gains and often degrading zero-shot transfer. We introduce RLBind, a two-stage adversarial-invariant cross-modal alignment framework for robust unified embeddings. Stage 1 performs unsupervised fine-tuning on clean-adversarial pairs to harden the visual encoder. Stage 2 leverages cross-modal correspondence by minimizing the discrepancy between clean/adversarial features and a text anchor, while enforcing class-wise distributional alignment across modalities. Extensive experiments on Image, Audio, Thermal, and Video data show that RLBind consistently outperforms the LanguageBind backbone and standard fine-tuning baselines in both clean accuracy and norm-bounded adversarial robustness. By improving resilience without sacrificing generalization, RLBind provides a practical path toward safer multi-sensor perception stacks for embodied robots in navigation, manipulation, and other autonomy settings.
title RLBind: Adversarial-Invariant Cross-Modal Alignment for Unified Robust Embeddings
topic Robotics
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2509.14383