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

Cheng Zhang; Qing Cai; Xingzheng Wu; Xun Yang; Xiaojun Chang; Bingkun Bao; Liqiang Nie; Xinwang Liu

cs.AIApr 5, 2026

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

Authors: Cheng Zhang, Qing Cai, Xingzheng Wu, Xun Yang, Xiaojun Chang, Bingkun Bao, Liqiang Nie, Xinwang Liu,

Organizations: School of Information Science and Engineering, Ocean University of China, Qingdao, Shandong 266100, China · Innovation School of Artificial Intelligence, Hefei University of Technology, Hefei 230009, China · School of Information Science and Technology, University of Science and Technology of China, Hefei 230026, China · School of Computer Science and Information Engineering, Hefei University of Technology, Hefei 230009, China ·

Abstract

Foundation models have demonstrated impressive performance in enhancing healthcare efficiency across a wide range of medical applications. Nevertheless, their limited ability to perceive, understand, and interact with the physical world significantly constrains their effectiveness in real-world clinical workflows, where safety-critical decision-making and physical execution are tightly coupled. Recently, embodied artificial intelligence (AI) has emerged as a promising physical-interactive paradigm for intelligent healthcare, enabling agents to operate in complex medical environments. As research in this area rapidly expands, understanding how intelligent agents function as integrated, end-to-end systems in clinical environments becomes increasingly critical. However, existing surveys on medical embodied AI largely emphasize individual aspects or functional components, lacking a unified system-level organization of the field. To support and consolidate recent advances, we systematically survey the core components of medical embodied AI, with a particular emphasis on the coordinated integration of perception, decision-making, and action. We further review representative medical applications and relevant datasets, and we analyze the major challenges encountered in real-world clinical practice. Finally, we discuss key directions for future research in this rapidly evolving field. The associated project can be found at https://github.com/VMVLab/Medical_Embodied_AI_Paper_List.

Figures & tables

Fig. 1: Overview of a medical embodied artificial intelligence framework. Medical embodied agents interact with simulated and real clinical environments via a perception–decision–action loop.

Fig. 2: Overview structure of this survey.

Fig. 3: Conceptual foundations of embodied AI and its relevance to medical embodied intelligence. a, Publication trends, temporal evolution over the past decade, and representative keywords of embodied AI based on Google Scholar statistics. b, Four developmental stages of embodied AI: conceptual germination, paradigm shift, learning-driven, and large-model–empowered stages. c, Comparison between disembodied intelligence and embodied AI, highlighting the latter’s ability to interact with the environment. d, Core components of embodied AI, including agents and environments at the system level, and embodied perception, decision-making, and action at the technical level.

Components	Function	Sub-Direction
Embodied Perception	Provides multimodal understanding of the environment.	Object Perception
		Scene Perception
		Behavior Perception
		Expression Perception
Embodied Decision-Making	Converts perception into adaptive strategies.	Task Planning
		Embodied Navigation

TABLE I: Overview of the core components, their respective functions, and sub-directions in embodied AI.

Fig. 4: Overview of medical embodied AI and its hierarchical organization with representative methods.

Fig. 5: Overview of medical embodied perception, including medical instrument and organ recognition, surgical and clinical environment perception and modeling, medical operation behavior detection, and emotional interaction understanding.

Fig. 6: Overview of medical embodied decision-making, including medical workflow modeling and task planning, medical navigation systems, and clinical question-answering and decision-support mechanisms.

Fig. 7: Overview of medical embodied action, including medical imitation-based action, medical reinforcement-based action, and medical large-model-driven action.

Category	System/Robot	Setting	Core functions	Autonomy
Surgical Robot	da Vinci [ 121 ]	OR	Surgeon-controlled manipulation; improves precision and consistency	Teleop
	EMARO [ 122 ]	OR	Endoscope stabilization and positioning; reduces assistant burden	Shared
	ROSA [ 123 ]	OR	Image-guided navigation and positioning; improves placement accuracy	Assisted
	PRECEYES [ 124 ]	OR	Micron-level manipulation; tremor suppression	Teleop
	ROBODOC [ 125 ]	OR	High-precision bone preparation for improved implant fit	Supervised
	Symani [ 126 ]	OR	Enhanced micro-scale dexterity and consistency	Teleop

TABLE II: Integrated application systems of medical embodied AI.

Fig. 8: Intelligent Caregiving and Companion Robot. (a) Pepper [ 133 ] . (b) Arash [ 135 ] .(c) Giraff [ 137 ] . (d) AIBO [ 132 ] . (e) PARO [ 131 ] . (f) Robear [ 136 ] . (g) ElliQ [ 134 ] . (h) Telenoid [ 138 ] .

Fig. 9: The datasets of medical embodied AI, which are categorized into perception, decision-making, action, and simulation & synthetic.

Platform	Representative Tasks	Learning Paradigms	Strength	Limitation
SurRoL [ 196 ]	Grasping, cutting, suturing, general robotic surgery	Imitation learning; reinforcement learning	Strong physical realism; diverse surgical tasks	Moderate visual fidelity; high compute cost
ORBIT-Surgical [ 197 ]	Surgical dexterity learning; active perception	Reinforcement learning; imitation learning	Excellent visual realism; efficient GPU-parallel simulation	High hardware demand; complex system setup
Surgical Gym [ 198 ]	Massive RL training; rapid policy iteration	Reinforcement learning	Extremely high training speed; scalable optimization	Lower physical and visual realism
LapGym [ 111 ]	Laparoscopic manipulation; path planning; human-in-the-loop control	Imitation learning; reinforcement learning	Highly extensible; multimodal sensing support	Limited task diversity; user-dependent fidelity
SonoGym [ 199 ]	Ultrasound navigation; bone reconstruction; intervention planning	Reinforcement learning; imitation learning	High anatomical realism for ultrasound procedures	Modality-specific; limited generality

TABLE III: Representative simulation platforms for medical embodied AI.

Healthcare Robotics

See all 24 papers

Safety in Embodied AI: A Survey of Risks, Attacks, and Defenses
Mar 28, 2026Xiao Li, Xiang Zheng, Yifeng Gao +35Embodied AI AI Safety
Fudan University · City University of Hong Kong · Jilin University +10
HealthAgentBench: A Unified Benchmark Suite of Realistic Agentic Healthcare Environments for Challenging Frontier AI Agents
Jun 30, 2026Qianchu Liu, Sheng Zhang, Guanghui Qin +16Long-Horizon Agent Evaluation AI Agent Benchmarks
Microsoft Research
Meissa: Multi-modal Medical Agentic Intelligence
Mar 9, 2026Yixiong Chen, Xinyi Bai, Yue Pan +2Medical LLMs Multimodal Agents
Johns Hopkins University, Baltimore, MD, USA · Cornell University, Ithaca, NY, USA

Mar 28, 2026J/K move · Enter open · S save

Safety in Embodied AI: A Survey of Risks, Attacks, and Defenses

Xiao Li, Xiang Zheng, Yifeng Gao, Xinyu Xia+34

Fudan University · City University of Hong Kong · Jilin University+10

Embodied AI AI Safety AI Agent Security Robot Safety Human-Robot Interaction

Open paper AskPDF arXiv

Embodied Artificial Intelligence (Embodied AI) integrates perception, cognition, planning, and interaction into agents that operate in open-world, safety-critical environments. As these systems gain autonomy and enter domains such as transportation, healthcare, and industrial or assistive robotics, ensuring their safety becomes both technically challenging and socially indispensable. Unlike digital AI systems, embodied agents must act under uncertain sensing, incomplete knowledge, and dynamic human-robot interactions, where failures can directly lead to physical harm. This survey provides a comprehensive and structured review of safety research in embodied AI, examining attacks and defenses across the full embodied pipeline, from perception and cognition to planning, action and interaction, and agentic system. We introduce a multi-level taxonomy that unifies fragmented lines of work and connects embodied-specific safety findings with broader advances in vision, language, and multimodal foundation models. Our review synthesizes insights from over 500 papers spanning adversarial, backdoor, jailbreak, and hardware-level attacks; attack detection, safe training and robust inference; and risk-aware human-agent interaction. This analysis reveals several overlooked challenges, including the fragility of multimodal perception fusion, the instability of planning under jailbreak attacks, and the trustworthiness of human-agent interaction in open-ended scenarios. By organizing the field into a coherent framework and identifying critical research gaps, this survey provides a roadmap for building embodied agents that are not only capable and autonomous but also safe, robust, and reliable in real-world deployment.

Abstract

Healthcare Robotics

Explore similar work

Safety in Embodied AI: A Survey of Risks, Attacks, and Defenses

HealthAgentBench: A Unified Benchmark Suite of Realistic Agentic Healthcare Environments for Challenging Frontier AI Agents

Meissa: Multi-modal Medical Agentic Intelligence