I am currently pursuing the Ph.D. degree, advised by Prof. Kai Liu in computer science at Chongqing University, Chongqing, China. My research interests include vehicular networks, edge computing, and deep reinforcement learning. I have published more than 10 papers with total google scholar .

πŸ”₯ News

  • 2020.10: πŸŽ‰πŸŽ‰ One paper is accepted by JSA!
  • 2022.10: πŸŽ‰πŸŽ‰ My google scholar citations have reached 100!
  • 2022.09: One invention patent is granted!
  • 2022.06: One paper is accepted by IEEE ITSC 2022!

πŸ•’ Research


Enabling Digital Twin in Vehicular Edge Computing: A Multi-Agent Multi-Objective Deep Reinforcement Learning Solution
Xincao Xu, Kai Liu, Penglin Dai, and Biwen Chen*

  • We present a DT-VEC architecture, where the heterogeneous information can be sensed by vehicles and uploaded to the edge node via V2I communications. The DT-VEC are modeled at the edge node, forming a logical view to reflect the physical vehicular environment.
  • We model the DT-VEC by deriving an ISAC-assisted sensing model and a reliability-guaranteed uploading model.
  • We formulate the bi-objective problem to maximize the system quality and minimize the system cost, simultaneously. In particular, we define the quality of DT-VEC by considering the timeliness and consistency, and define the cost of DT-VEC by considering the redundancy, sensing cost, and transmission cost.
  • We propose a multi-agent multi-objective (MAMO) deep reinforcement learning solution implemented distributedly in the vehicles and the edge nodes. Specifically, a dueling critic network is proposed to evaluate the advantage of action over the average of random actions.
  • Submitted to IEEE Transactions on Consumer Electronics (under review)
System Model
System Model

Cooperative Sensing and Heterogeneous Information Fusion in VCPS: A Multi-agent Deep Reinforcement Learning Approach
Xincao Xu, Kai Liu*, Penglin Dai, Ruitao Xie, and Jiangtao Luo

  • We present a cooperative sensing and heterogeneous information fusion architecture in VCPS via vehicular edge computing. The heterogeneous information can be sensed via either onboard sensors such as LIDAR, GPS, and cameras, or roadside infrastructures such as traffic lights. The sensed information is queued in vehicles for uploading via the V2I bandwidth, which is allocated by the corresponding edge node. Logical views can be constructed via the information fusion at edge nodes, and different views may be required to enable upper-layer applications.
  • We formulate the problem to maximize the quality of VCPS. Specifically, we derive a cooperative sensing model, in which the information queuing and data uploading are modeled based on the multi-class M/G/1 priority queue and the Shannon theory, respectively. Then, we derive a heterogeneous information fusion model by modeling the timeliness, completeness, and consistency of views. On this basis, a new metric called Age of View (AoV) is defined to quantitatively measure the quality of information fusion. Finally, we model the quality of VCPS and present the optimization objective, which is to maximize the VCPS quality.
  • We propose a multi-agent deep reinforcement learning solution. Specifically, the solution model is presented, in which vehicles act as independent agents with action space of determining the sensing frequencies and uploading priorities, and the edge action space is the V2I bandwidth allocation. The system state consists of vehicle sensed information, edge cached information, and view requirements. The system reward is defined as the achieved VCPS quality. In particular, the system reward is divided into the Difference Reward (DR) to capture vehicle individual contributions on view constructions by the DR-based credit assignment.
  • Submitted to IEEE Transactions on Intelligent Transportation Systems (under review)
NOMA-based VEC
NOMA-based VEC

Joint Task Offloading and Resource Optimization in NOMA-based Vehicular Edge Computing: A Game-Theoretic DRL Approach
Xincao Xu, Kai Liu*, Penglin Dai, Feiyu Jin, Hualing Ren, Choujun Zhan, and Songtao Guo

  • We present a NOMA-based VEC architecture, where the vehicles share the same frequency of bandwidth resources and communicate with the edge node with the allocated transmission power. The tasks arrive stochastically at vehicles and are heterogeneous regarding computation resource requirements and deadlines, which are uploaded by vehicles via V2I communications. Then, the edge nodes with heterogeneous computation capabilities, i.e., CPU clock speed, can either execute the tasks locally with allocated computation resources or offload the tasks to neighboring edge nodes through a wired network.
  • We propose a cooperative resources optimization (CRO) problem by jointly offloading tasks and allocating communication and computation resources to maximize the service ratio, which is the number of tasks serviced before the deadlines divided by the number of requested tasks. Specifically, a V2I transmission model considering both intra-edge and inter-edge interference and a task offloading model considering the heterogeneous resources and cooperation of edge nodes are theoretically modeled, respectively.
  • We decompose the CRO problem into two subproblems: 1) task offloading game model and 2) resource allocation convex problem. Specifically, we model the first subproblem as a non-cooperative game among edge nodes, which is further proved as an EPG with the existence and convergence of NE. Then, we design a MAD4PG algorithm, a multi-agent version of D4PG, to achieve the NE, where edge nodes act as independent agents to determine the task offloading decisions and receive the achieved potential as rewards. Further, we model the second subproblem as two independent convex problems and derive an optimal mathematical solution based on the gradient-based iterative method and KKT condition.
  • Accepted by Journal of Systems Architecture [JCR Q1|SCI Q2|CCF B]

πŸ’» Publications

JCR: Journal Citation Reports by Clarivate Com.
SCI: Journal Partition List by National Science Library, Chinese Academy of Sciences
CCF: Recommended Publications by China Computer Federation
*: Corresponding Author



Chinese Papers

  • Xincao Xu, Kai Liu*, Chunhui Liu, Hao Jiang, Songtao Guo and Weiwei Wu, Potential Game Based Channel Allocation for Vehicular Edge Computing, Tien Tzu Hsueh Pao/Acta Electronica Sinica, volume 49, issue 5, pp.851-860, July 2021. [CCF T1]
  • Xincao Xu, Yi Zhou, Kai Liu, Chaocen Xiang, Yantao Li and Songtao Guo, Potenial Game based Distributed Channel Allocation in Vehicular Fog Computing Environments, 14th China Conference on Internet of Things (CWSN’20), Dunhuang, China, September 18-21, 2020. (Best Paper Candidate)

✨ Others

πŸ“„ Patents

  • Xincao Xu, Kai Liu, Dong Li, A Control Plane View Construction Method for Software-Defined Vehicular Networks, Chinese Invention Patent (ZL202110591822.1), 2022.
  • Liu Kai, Zhang Lang, Xincao Xu, Ren Hualing, Zhou Yi, An Edge Computing Based Collision Warning Method for Vehicles in Blind Areas, Chinese Invention Patent (ZL201910418745.2), 2021.
  • Ren Hualing, Liu Kai, Chen Mengliang, Zhou Yi, Xincao Xu, A Fog Computing-based Information Acquisition, Computing, and Transmission Architecture, Chinese Invention Patent (ZL201910146357.3), 2021.

🚧 Grants

  • General Project of National Natural Science Foundation of China, Computing Model Deployment and Collaborative Cross Domain Optimization for Edge Intelligence of Internet of Vehicles, 62172064, 2022/01–2025/12. (Participation)
  • General Project of National Natural Science Foundation of China, Research on High Performance Data Transmission and Reception Mechanism of Wireless Energy-capable Communication Network Based on Successive Interference Cancellation, 62072064, 2021/01-2024/12. (Participation)
  • General Project of National Natural Science Foundation of China, Research on Architecture and Protocols for Large-scale Data Services in Converged Heterogeneous Internet of Vehicles, 61872049, 2019/01–2022/12. (Participation)
  • General Project of National Natural Science Foundation of China, Research on Multi-Channel I2V/V2V Hybrid Communication and Temporal Information Service in Internet of Vehicles, 61572088, 2016/01–2019/12. (Participation)

πŸ“Ί Demos

  • Non-Light-of-Sight Collision Warning System

πŸ“– Educations

  • 2017.09 - 2023.06 (now), Ph.D, Computer Science, College of Computer Science, Chongqing University, Chongqing. (Successive Master-Doctor Program)
  • 2013.09 - 2017.06, Bachelor, Network Engineering, College of Computer and Control Engineering, North Univeristy of China, Taiyuan.

πŸŽ’ Learning


  • Introduction to Algorithms by MIT [Link]
  • Meachine Learning by Prof. Hung-yi Lee [Link]
  • Meachine Learning by Prof. Richard Xu [Link]
  • Deep Learning by Prof. Andrew Ng [Link]
  • Deep Reinforcement Learning by Dr. Shusen Wang [Link]


  • Meachine Learning by Prof. Zhi-Hua Zhou [Chinese]
  • Deep Learning by Dr. Ian Goodfellow, Prof. Yoshua Bengio, and Prof. Aaron Courville [English] [Chinese]
  • Machine Learning Yearning by Prof. Andrew Ng [Chinese]
  • Dive into Deep Learning by Dr. Aston Zhang, Dr. Zachary C. Lipton, Dr. Mu Li, and Dr. Alexander I. Smola [English] [Chinese]


  • Google Scholar: Largest Search Engine for Academic Publishing [Link]
  • dblp: Computer Science Bibliography [Link]
  • Conference Partner: Information about Academic Conferences [Link]
  • LetPub: Information about SCI Journals [Link]
  • Academic Accelerator: Wonderful Tool to Accelerate Your Scientific Research [Link]
  • WikiCFP: A Wiki for Calls for Papers [Link]
  • Zotero: Personal Research Assistant [Link]