Cephalopod-inspired embodied camouflage system for simultaneous evasion of artificial and human visual perception

Differences in visual perception mechanisms between biological and artificial intelligence make it difficult to design camouflage that is both visually natural and effective against modern detectors. Existing adversarial approaches often create conspicuous patterns and are typically static, limiting their use in dynamic real-world scenarios. Here we present CephCam, an embodied closed-loop camouflage system that integrates environment-aware color extraction, optimization-based texture generation that jointly balances visual naturalness and detector evasion, and angle-aware display compensation for physical deployment. This system involves an eye-brain-skin framework of sensing, computation, and display, motivated by the naturally evolved cephalopod camouflage, but implemented with specialized cameras, computing modules, and OLED screens on a model vehicle. CephCam targets simultaneous evasion of both human observers and modern object detectors across diverse digital and physical environments, significantly outperforming strong baselines targeting either alone. These findings underscore shared vulnerabilities in human and artificial visual processing systems, showcasing CephCam's potential for robust, adaptive, and embodied camouflage applications.