A blueprint for robotic surgery: From algorithmic control to embodied intelligence

Surgical robotics is at a pivotal transition, moving from master-slave teleoperation towards intelligent systems capable of autonomous task execution. The ultimate vision is not to replace the surgeon, but to create a collaborative companion that augments human expertise by handling repetitive and fatiguing sub-tasks. The most profound challenge lies in bridging the “intelligence chasm”—encoding the nuanced, experience-based “common sense” that defines surgical skill, which is largely absent from procedural manuals. This perspective outlines key research frontiers to address this challenge. Hierarchical, language-conditioned frameworks, such as the surgical robot transformer (SRT-H), are proving effective for long-horizon autonomy, using language for high-level planning and real-time self-correction to achieve high levels of autonomy in realistic ex vivo procedures. To overcome the field’s profound data scarcity, the concept of surgical embodied intelligence offers a transformative solution. By leveraging high-fidelity simulators (e.g., SurRoL) and hybrid AI paradigms, policies can be trained at scale and transferred with zero-shot success to real-world robots, demonstrating generalizable skills in both ex vivo and in vivo settings. These architectures often use general foundation models for a perceptual “warm start”, refined by domain-specific middle-layer models to ensure safety and reliability. The future of surgical autonomy hinges on fostering a seamless human-AI symbiosis through interactive, explainable systems and leveraging these intelligent platforms to revolutionize surgical training with AI-driven guidance. This integrated approach, supported by parallel advancements in safety verification and regulatory standards, will shape the next generation of surgery as a collaborative synergy between human intuition and artificial intelligence.