A Survey on Cloud-Edge-Terminal Collaborative Intelligence in AIoT Networks
Authors: Jiaqi Wu, Jing Liu, Yang Liu, Lixu Wang, Zehua Wang, Wei Chen, Zijian Tian, Richard Yu, Victor C. M. Leung
Organizations: School of Artificial Intelligence, China University of Mining and Technology (Beijing), Beijing 100083, China · Department of Electrical and Computer Engineering, University of British Columbia, Vancouver, BC V6T 1Z4, Canada · Division of Natural and Applied Sciences, Duke Kunshan University, Kunshan 215316, China · Department of Electrical and Computer Engineering, The University of British Columbia, Vancouver, BC V6T 1Z4, Canada · School of Information Science and Technology, Fudan University, Shanghai 200433, China · Department of Computer Science, The University of Toronto, ON M5S 1A1, Canada · Department of Electrical and Computer Engineering, The University of British Columbia, Vancouver, 2332 Main Mall Vancouver, BC Canada V6T 1Z4 · School of Computer Science and Technology, China University of Mining and Technology, Xuzhou 221116, Jiangsu province, China · Department of Systems and Computer Engineering, Carleton University, Ottawa, ON K1S 5B6, Canada · Artificial Intelligence Research Institute, Shenzhen MSU-BIT University, Shenzhen 518172, China · College of Computer Science and Software Engineering, Shenzhen University, Shenzhen 528060, China
The proliferation of Internet of things (IoT) devices in smart cities, transportation, healthcare, and industrial applications, coupled with the explosive growth of AI-driven services, has increased demands for efficient distributed computing architectures and networks, driving cloud-edge-terminal collaborative intelligence (CETCI) as a fundamental paradigm within the artificial intelligence of things (AIoT) community. With advancements in deep learning, large language models (LLMs), and edge computing, CETCI has made significant progress with emerging AIoT applications, moving beyond isolated layer optimization to deployable collaborative intelligence systems for AIoT (CISAIOT), a practical research focus in AI, distributed computing, and communications. This survey describes foundational architectures, enabling technologies, and scenarios of CETCI paradigms, offering a tutorial-style review for CISAIOT beginners. We systematically analyze architectural components spanning cloud, edge, and terminal layers, examining core technologies including network virtualization, container orchestration, and software-defined networking, while presenting categorizations of collaboration paradigms that cover task offloading, resource allocation, and optimization across heterogeneous infrastructures. Furthermore, we explain intelligent collaboration learning frameworks by reviewing advances in federated learning, distributed deep learning, edge-cloud model evolution, and reinforcement learning-based methods. Finally, we discuss challenges (e.g., scalability, heterogeneity, interoperability) and future trends (e.g., 6G+, agents, quantum computing, digital twin), highlighting how integration of distributed computing and communication can address open issues and guide development of robust, efficient, and secure collaborative AIoT systems.
Figures & tables
Fig. 1: Research trends in computing paradigms and AIoT from 2021 to 2025, measured by publications and citations.
Ref.
Year
Domain
Resource Management
Model Evolution
Security Aspects
Computing Paradigms
Applications
Task Off.
Res. Alloc.
Model Comp.
Know. Trans.
Security
Privacy
Edge
Cloud
ECC
CETC
AIoT
Manuf.
Transport.
Health
Agri.
[ 65 ]
2019
Edge-Cloud Systems, Application Mgmt.
●
●
◐
◐
◐
○
●
●
◐
○
◐
◐
◐
◐
○
[ 96 ]
2019
Network Architecture, Resource Allocation
●
●
○
○
○
○
●
◐
◐
○
◐
◐
◐
◐
◐
[ 204 ]
2019
IoT, Distributed Systems
●
●
○
◐
●
●
●
●
●
◐
●
◐
◐
◐
◐
[ 18 ]
2020
Resource optimization, AI algorithms
●
●
○
◐
○
○
●
●
◐
○
◐
◐
◐
○
○
[ 233 ]
2022
Metaverse, Digital Twin
●
●
○
○
○
◐
●
◐
●
○
◐
○
◐
○
○
TABLE I: Comparison of recent surveys on cloud-edge-terminal collaboration and related AIoT systems, evaluating how each survey covers resource management, model evolution, security aspects, computing paradigms, and application domains.
Fig. 2: Illustrative structure of cloud-edge-terminal collaborative intelligence (CETCI) in AIoT networks, highlighting the end-to-end survey scope that connects architectural layers, enabling technologies, intelligent management, and collaborative learning within a single cross-layer framework.
Fig. 3: Topology diagram of CETCI research scope and supporting components, showing how deployability is shaped by the coupling among resource management, collaborative learning, security and privacy protection, and data management.
Acronym
Definition
AI
Artificial intelligence
AIoT
Artificial intelligence of things
CETCI
Cloud-edge-terminal collaborative intelligence
CISAIoT
Collaborative intelligence systems for AIoT
DDL
Distributed deep learning
DP
Differential privacy
TABLE II: List of key acronyms used in this survey.
Metric
Focus in AIoT
Key Optimization Goal
Response Time
Inference latency, end-to-end delay
minimize total latency ( Ltotal ) via optimal offloading
Energy Consumption
Battery life, green computing
minimize total energy ( Etotal ) across layers
Resource Utilization
CPU/GPU/Bandwidth efficiency
maximize utilization ( Utotal ) without congestion
Quality of Service (QoS)
Reliability, throughput, availability
maintain QoS within feasible region ( Cfeasible )
Quality of Experience
User satisfaction, perceived quality
maximize subjective satisfaction ( Qexp )
TABLE III: Key Performance Metrics in CETCI Systems.
Fig. 4: Overview of performance metrics for evaluation.
Method
Key Tech.
Metrics
Limitations
CETC
TO
Binh et al. [ 29 ]
Reinforcement Learning (RL)
Tolerance Time
Convergence speed dependent on state space size.
✗
✓
ADPRL [ 47 ]
Deep RL (DRL)
Task Completion Time, Energy Consumption
High training overhead and requires extensive data.
TABLE IV: Summary of learning-based and optimization-based task offloading strategies in CETCI AIoT systems.
Method
Key Tech.
Metrics
Pros & Cons
CETC
RA
OTFAC [ 76 ]
Federated Learning + Actor-Critic
Delay, Energy Consumption
Adaptive and communication-efficient; complex to implement.
✗
✓
Fang et al. [ 69 ]
Deep Reinforcement Learning
Latency, Resource Utilization
Handles dynamic environments well; can have slow convergence.
✗
✓
Huang et al. [ 101 ]
DRL with Hybrid Action Space
Energy Consumption, Latency
Enables joint optimization; high training cost.
✗
✓
DRLS-VNE [ 285 ]
DRL
Security, Resource Utilization
Security-aware approach; state space can be very large.
✓
✓
SA-BPSO [ 48 ]
Hybrid Swarm Optimization
Total User Overhead
Effective heuristic; performance is sensitive to parameters.
✗
✓
Hua et al. [ 100 ]
Joint Optimization (Power, Scheduling)
Energy Consumption
Energy-efficient design; assumes accurate state information.
✗
✓
TABLE V: Summary of resource allocation techniques.
Fig. 6: Different types of collaboration learning paradigms.
Method
Key Tech.
Optimization Objectives
Pros & Cons
ECC
CETC
PECo [ 56 ]
Point Cloud Compression
Geometric Accuracy, Structural Compactness
Efficient for 3D data; specialized use case.
✗
✗
CDC [ 58 ]
Collaborative Compression
Bandwidth Efficiency, Accuracy
Reduces data transmission; requires coordination.
✓
✗
Hawlader et al. [ 90 ]
Knowledge Distillation
Detection Quality, Latency
Improves edge model accuracy; teacher dependent.
✓
✗
BPS [ 98 ]
Model Partitioning
Throughput, Latency
Balances load; sensitive to partition point.
✓
✓
Joshi et al. [ 108 ]
Incremental Learning + DT
Energy, Latency, Accuracy
Adapts to changes; high DT maintenance cost.
✓
✓
CO-PILOT [ 117 ]
Split Computing
End-to-end Latency, Accuracy
Lowers device computation; network sensitive.
✓
✗
TABLE VI: Summary of model compression and knowledge distillation techniques for CETCI AIoT systems, with emphasis on update cost, edge deployability, and sensitivity to teacher, partition, or synchronization choices.
Method
Learning Paradigm
Metrics
Pros & Cons
Zhu et al. [ 299 ]
MARL
Average Age of Information
Decentralized and privacy-preserving; high communication overhead.
Park et al. [ 191 ]
MADRL
Quality of Service
Scalable for large agent teams; requires centralized controller.
DRL-CCMCO [ 45 ]
Deep RL
Execution Delay, Energy Consumption
Fast convergence and stable; sensitive to hyperparameters.
Simple implementation; struggles with large state spaces.
Coop-MADRL [ 223 ]
Cooperative MADRL
Network Utilization, Task Latency
Promotes cooperation; credit assignment is challenging.
TABLE VII: Summary of RL and MARL approaches, emphasizing their convergence sensitivity, coordination overhead, and suitability for decentralized CETCI control.
Method
Scenarios
Performance Metrics
ECC
CETC
Yang et al. [ 266 ]
AIoT Data Prediction
Privacy and Accuracy
✓
✓
Rahman et al. [ 200 ]
Industry 4.0 CPS
Security, Detection Performance
✓
✗
Fang et al. [ 70 ]
CETC-AIoT
Computational Complexity
✓
✓
Chen et al. [ 42 ]
Workflow Scheduling
Execution Cost, Deadline Constraints
✓
✗
Wang et al. [ 247 ]
Security Inference
Privacy-Utility Tradeoff
✓
✗
Zhou et al. [ 295 ]
AIoT Services
QoS, Effectiveness, Efficiency
✓
✗
TABLE VIII: Summary of methods addressing security threats and vulnerabilities in CISAIoT networks.
Method
Type
Scenarios
Performance Metrics
ECC
CETC
Raj et al. [ 201 ]
Edge MLOps
AIoT Fleet Analytics
Forecasting Accuracy
✓
✗
Becker et al. [ 27 ]
AIOps
Edge Computing
Resource Utilization, Detection Performance
✗
✗
QuAsyncFL [ 159 ]
Federated Learning
Cloud-Edge-Terminal AIoT
Communication Efficiency, Convergence
✓
✓
Underwood et al. [ 232 ]
Data Compression
ML/AI Training
Compression Ratio, Model Quality
✗
✗
Noura et al. [ 186 ]
DL-based Compression
Multimedia IoT
Visual Quality, Compression Ratio
✓
✗
Edge Source Coding [ 165 ]
Lossless Compression
Edge Networks
Number of Bits Needed
✗
✗
TABLE IX: Summary of data preprocessing and management for transmission in resource-constrained environments.
Fig. 11: Illustration of CETCI in smart healthcare.
Fig. 12: Overview of future research directions.
Appendix figures & tables13 assets
Supplementary material from the paper’s appendix.
Appendix
Method
Year
Venue
Algorithm Type
Scenarios
Metrics
Performance Results
TIGO [ 93 ]
2024
IEEE TSC
Two-stage Iterated Greedy Optimization
Cloud-Edge Collaborative Computing
Average Response Time
Better response times in less time compared to existing methods
DSCF [ 118 ]
2024
IEEE IoTJ
CMDP-based Framework
Distributed Serverless Edge Clouds
Inference Latency
Less than half the inference latency compared to local computing
AWSH [ 121 ]
2024
IEEE TMC
Hyper-Heuristic Algorithm
5G Mobile Edge Clouds
Latency, Load Balancing
Superior performance compared to SOTA solutions
Lu et al. [ 166 ]
2024
IEEE TVT
Multi-Agent Reinforcement Learning
IoV Edge Cloud
Quality of Service
Effective performance in complex network situations
Nguyen et al. [ 180 ]
2024
IEEE TCC
Alternating Optimization
Space-Air-Ground Networks
Energy Consumption, Delay
Superior effectiveness compared to baselines
Qiu et al. [ 196 ]
2024
IEEE TMC
Online Algorithm
Edge Intelligence Cloud
Security, Reliability, Throughput
Promising performance compared to baseline algorithms
Edge-cloud computing distributes data processing and AI workloads between nearby devices and remote cloud services to balance latency, cost, energy, and scale.
We are moving from an information age to the age of intelligence. A decade, or possibly less than that, data will not be the gold anymore rather the derived intelligence out of the data and the information we posses from the edge of the network. Existing Edge Intelligence research focuses mainly on two directions: using AI for edge resource management and deploying lightweight AI models on edge devices. However, existing edge computing research lacks an intelligence-centric framework in which derived intelligence is treated as a first-class, independently manageable entity that can be described, discovered, observed, shared, reused, and dynamically clustered across heterogeneous edge devices and applications. To address these research gaps, we introduced Clustered Edge Intelligence, a visionary intelligence-centric approach. The aim of CEI is to make intelligence a shareable and reusable first-class entity that can be independently represented, discovered, observed, exchanged, and managed across the distributed edge-cloud continuum. We present a three layer CEI architecture and examine enabling technologies and research dimensions, including intelligence inventories, semantic knowledge representation, communication, discoverability, observability, lifecycle automation, clustering mechanisms, marketplaces, interoperability, and standardization.
Chinmaya Kumar Dehury, Boris Sedlak, Alaa Saleh +4
Department of PComputer Science, IISER Berhampur, Odisha, India · Distributed Systems Group, TU Wien, 1040 Vienna, Austria. · Department of Computer Science, University of Helsinki, 00100 Helsinki, Finland +4
The integration of AI into Internet of Things (AIoT) systems has gradually transformed them from passive data collection infrastructures into intelligent systems capable of anomaly detection, predictive maintenance, classification, forecasting, and optimization. However, most existing solutions still rely on task-specific models that infer from sensor data; thus, system-wide capabilities such as real-time reasoning, adaptive planning, autonomous coordination, learning, tool use, and contextual decision-making remain limited. This paper examines Agentic IoT as a next-generation cognitive IoT paradigm that integrates the perception, reasoning, planning, learning, and action capabilities of autonomous AI agents with cyber-physical systems. Agentic IoT aims to transform IoT from data-centric sensing and inference infrastructures into distributed cognitive agent ecosystems operating across the device/edge-fog-cloud continuum. The paper first grounds this transition as a paradigm shift and positions Agentic IoT in relation to AIoT, edge intelligence, multi-agent systems, and the Internet of Agents. It then systematically reviews current studies, presents a holistic architectural framework, discusses domain-specific application potential, and identifies key technical, operational, and research challenges together with future research directions.
Rümeysa Hilal Sevinç, Bahaeddin Türkoğlu, İbrahim Kök
Department of Artificial Intelligence and Data Engineering, Ankara University, Ankara, Türkiye
The convergence of decentralized architectures integrating Machine Learning, Computer Vision and Low Power Wide Area Networks is increasingly becoming an integral part of our daily existence. Internet of Things serves as a real-time data conduit enhancing decision making via embedded technology and continuous data exchange. This paper explores the feasibility of Edge Computing as a foundational pillar in this evolving landscape. We experiment under real world, dynamic conditions, evaluate the technological aspects, strategies, process flows and key observations under the broad Edge Computing domain. Research pathways include Multi-access Edge topologies in future 6G networks, model quantization, and satellite-enhanced communication platforms. Additionally, a discussion is added supporting the advanced AI functionalities, including zero-shot learning, multi modal perception, and decentralized generative AI, thereby broadening the scope of intelligent applications across various domains. The significance and research objective of this study are threefold: (1) evaluation of LoRaWAN and satellite IoT communication strategies, (2) analysis of CV workloads on edge hardware and (3) future research directions where Edge Computing can support low-latency, energy-efficient and socially impactful IoT applications. By explicitly addressing these aspects, we aim to establish a clear link between the technological feasibility, ultimately with a practical and socioeconomic relevance.
Panagiotis Savvidis, George A. Papakostas
Department of Computer Science, Democritus University of Thrace, Kavala 65404, Greece