Towards Next-Generation Healthcare: A Survey of Medical Embodied AI for Perception, Decision-Making, and Action

A new survey calls for a unified, system-level approach to medical embodied artificial intelligence, arguing that integrating perception, decision-making, and action is critical for deploying AI in real-world clinical workflows [1]. The paper, submitted to arXiv on 5 April 2026, systematically reviews the core components of medical embodied AI [1]. It contends that while foundation models have improved healthcare efficiency, their limited ability to perceive and interact with the physical world restricts their use in safety-critical settings where decisions and physical execution are tightly coupled [1]. Embodied AI, by contrast, enables agents to operate in complex medical environments as integrated, end-to-end systems [1]. Artificial intelligence is broadly defined as the capability of computational systems to perform tasks associated with human intelligence, including perception and decision-making, and to take actions that maximize the chances of achieving defined goals [2]. The field was founded as an academic discipline at a Dartmouth College workshop in 1956, and has since experienced cycles of optimism and so-called “AI winters” when funding and interest receded [3]. Investment boomed in the 2020s, fueled by advances in deep learning and the transformer architecture introduced in 2017, which enabled generative AI applications and large language models [3]. The survey emphasizes that existing literature on medical embodied AI largely examines individual aspects or functional components, lacking a unified system-level organization [1]. To address this gap, the authors review representative medical applications, relevant datasets, and major challenges encountered in clinical practice [1]. The work appears on arXiv, an open-access repository of electronic preprints that, as of November 2024, receives about 24,000 submissions per month and has surpassed two million articles [8]. The paper’s abstract page features arXivLabs, a framework launched in 2020 that allows community collaborators to develop and share experimental tools directly on the site [6]. These tools include the Bibliographic Explorer, which displays citation networks, and the CORE Recommender, which surfaces related open-access papers from a global network of repositories [7]. arXivLabs collaborators have access only to minimal, anonymized user data, and any other use is prohibited without written consent from arXiv [6]. The framework is currently pausing new proposals while the development team focuses on modernizing arXiv’s infrastructure and moving systems to the cloud [5].