Embodied Intelligence in Surgical Robotics: From Perception-Action Loops to Cognitive Collaboration

Surgical robotics is transitioning from teleoperated master-slave systems toward embodied intelligent agents that actively perceive, reason, and act within complex surgical environments. Although contemporary robotic platforms significantly enhance human motor dexterity through motion scaling and tremor suppression, they remain limited in addressing perceptual uncertainty, soft-tissue variability, and the cognitive demands of intraoperative decision-making, resulting in heavy reliance on continuous human supervision. Embodied Artificial Intelligence (AI) offers a paradigm shift by treating intelligence as an emergent property of tightly coupled interactions among a robot's physical embodiment, multimodal sensing, control policies, and environment. This Review synthesizes recent advances in embodied surgical robotics across three foundational dimensions: active multimodal perception that integrates visual, tactile, and force sensing in closed perception-action loops; learning-based and neuro-symbolic cognition that combines imitation learning, reinforcement learning, and clinical priors for adaptive planning under uncertainty; and compliant actuation and embodied control that embed safety and robustness directly into mechanical design and low-level control. Building on these developments, we propose a unified Six-Layer Embodied Architecture comprising Perception, Cognition, Action, Safety, Interaction, and System Integration, which clarifies functional roles while preserving cross-layer coupling to enable incremental autonomy within verifiable safety envelopes and calibrated human-robot collaboration. We discuss representative algorithms, hardware designs, and benchmark systems for each layer, and outline key translational challenges, including Sim2Real transfer in deformable tissues, certification of learning-enabled systems, interpretability and trust, and hospital-scale integration, positioning embodied intelligence as a pathway toward safer, more adaptive, and collaborative surgical robotic systems.