Ambient Intelligence for Immersive Networked Systems (AIINS) Lab
Research Field
Cheng-Hsin Hsu is a Professor in the Department of Computer Science at National Tsing Hua University (NTHU). What distinguishes Cheng-Hsin from many other academics is his extensive experience in industrial R&D. Prior to joining NTHU, he accumulated extensive industrial R&D experience, serving as a Senior Research Scientist at Deutsche Telekom R&D Lab in Silicon Valley (USA), following his tenures at Motorola and Lucent. This unique background allows him to bridge the gap between theoretical algorithms and real-world system deployment, ensuring that research conducted in his group has high practical value and industrial relevance. His work is highly influential, evidenced by over 5,800 citations, multiple Best Paper Awards (e.g., IEEE RTAS, IEEE CloudCom, IEEE SMARTCOMP, ACM EMS, and ACM MMSys), and his service as an Associate Editor for ACM TOMM.
Dr. Hsu’s research focuses on the end-to-end pipelines for: (i) immersive VR and $360^{\circ}$ video streaming, (ii) dynamic point cloud and 3D Gaussian Splats (3DGS) compression, and (iii) AI-assisted edge computing. He is deeply committed to talent cultivation, leveraging his industry insights to guide students toward impactful, original research. His students frequently win international recognition, including the Qualcomm Innovation Fellowship, MSRA Fellowship Finalists, and Novatek Scholarship. Successful interns will work on cutting-edge problems in volumetric media and drone analytics, gaining the rigorous training needed for top-tier industrial or academic careers.
Led by Cheng-Hsin Hsu at NTHU, the Ambient Intelligence for Immersive Networked Systems (AIINS) Lab is a dynamic team of systems-oriented innovators dedicated to bridging the gap between theory and real-world deployment. Leveraging Cheng-Hsin’s extensive industrial R&D experience (Deutsche Telekom, Motorola, and Lucent), we build next-generation prototypes for immersive media and smart environments. Our active research spans 3D Gaussian Splatting (3DGS), dynamic point clouds, Cloud XR over 5G/6G, autonomous drone swarms, and Digital Twins. We don't just analyze theories; we build scalable systems that solve real-world problems.
The AIINS Lab operates as a global hub for multimedia research. We maintain long-term strategic partnerships with world-class groups at UCI, Rutgers, NUS, AAU, Aalto, Northeastern, and UiO. Our students consistently achieve top-tier recognition, highlighted by a recent "winning streak" of Best Paper Awards (ACM MMSys 2025, EMS 2024, MADiMa 2023) and the prestigious 2026 Qualcomm Innovation Fellowship. With members having presented their work in over 60 cities across 20 countries, AIINS offers a vibrant, international environment for students passionate about pushing the boundaries of AI and Multimedia Networking.
The AIINS Lab is actively pushing the boundaries of multimedia networking and AI. The following list outlines our current primary research directions where we are building next-generation prototypes. However, we value creativity above all else—interns are not limited to these specific topics and are warmly welcome to propose novel ideas or interdisciplinary projects that align with our systems-oriented vision.
- Dynamic 3D Gaussian Splatting (4DGS) Streaming: Building on our prior success in streaming static high-fidelity 3D scenes, we are now extending our framework to handle time-varying, dynamic content. Interns will develop AI-driven compression algorithms to optimize the transmission of moving 3D objects and build WebXR-based players for universal access.
- Real-Time Drone Swarm Coordination: We have previously developed optimization algorithms for offline drone trajectory planning. The next phase focuses on online, real-time path planning. Interns will design "Next-Best-View" algorithms that enable swarms of drones to collaboratively explore and reconstruct large-scale environments in real-time.
- Generative AI for Network Digital Twins: Following our development of a software-defined controller for synchronizing IoT devices, we are now scaling up to city-level infrastructure. Research will focus on integrating Generative AI to predict network failures and automate self-healing processes for massive-scale Digital Twins.
- Predictive Foveated Rendering for 6G Cloud XR: Our team has successfully implemented gaze-adaptive rendering to reduce VR bandwidth consumption. To further mitigate latency in wireless networks, interns will apply deep learning models (RNNs/Transformers) to predict user eye movement, enabling lag-free volumetric video streaming over next-gen 5G/6G networks.
- ACM SIGMM Test of Time Paper Award, ACM Multimedia Systems Conference (MMSys'25), 2025
- Best Paper Award, ACM Multimedia Systems Conference (MMSys'25), 2025
- Honorable Mention Associate Editor, ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM), 2025
- Best Paper Award, ACM SIGCOMM Workshop on Emerging Multimedia Systems (EMS'24), 2024
- Best Associate Editor, ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM), 2024
- Visiting Faculty Award, J. Yang & Family Foundation, University of California, Irvine (UCI), 2022-2024
- Best Paper Award, ACM International Workshop on Multimedia Assisted Dietary Management (MADiMa'23), 2023
- Honorable Mention Associate Editor, ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM), 2021
- AI 2000 Most Influential Scholar Honorable Mention in Multimedia, AMiner, 2021
- Best Paper Award, IEEE International Conference on Smart Computing (SMARTCOMP'20), 2020
- Outstanding Reviewer/Best Reviewer Awards: ACM Multimedia'20, IEEE ICME'20, IEEE NOMS'20
- Outstanding Scholar Award, Foundations for the Advancement of Outstanding Scholarship (FAOS), 2018-2023
- Best Paper Award, IEEE International Conference on Cloud Computing Technology and Science (CloudCom'17), 2017
- Best Paper Award, Asia-Pacific Network Operations and Management Symposium (APNOMS'16), 2016
- Best Associate Editor, ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM), 2016
- IEEE Senior Member, IEEE, since 2016
- New Faculty Research Award, CSEE College, National Tsing Hua University, 2014
- Excellent Junior Research Investigator Grant, National Science Council (NSC), 2013-2016
- TAOS Best Paper Award, IEEE Global Communications Conference (GLOBECOM'12), 2012
- Best Paper Award, IEEE Real-Time and Embedded Technology and Applications Symposium (RTAS'12), 2012
- Best Demo Award, ACM International Conference on Multimedia (Multimedia'08), 2008
- Ph.D. in Computing Science (2009) Simon Fraser University, Canada
- Advisor: Prof. Mohamed Hefeeda
- Thesis: Efficient Mobile Multimedia Streaming
- M.Eng. in Electrical and Computer Engineering (2003) University of Maryland, College Park, USA
- M.S. in Computer Science and Information Engineering (2000) National Chung Cheng University, Taiwan
- Advisor: Prof. Daniel J. Buehrer
- Thesis: Making Java Applications Run Remotely
- B.S. in Mathematics (1996) National Chung Cheng University, Taiwan
Job Description
We are seeking a highly motivated Research Intern for a collaborative research visit focused on tackling cutting-edge challenges in robotic teleoperation, motion prediction, and network latency compensation.
This role aims to bridge the gap between network-based predictive algorithms (e.g., lightweight time-series predictions for latency compensation) and compact, task-relevant environment awareness (e.g., ego-centric visual/spatial understanding).
Our goal is to build a "spatial-aware" teleoperation framework. While existing network compensation models effectively predict motion trajectories over jittery connections, they are often "environment-blind." Conversely, recent breakthroughs in ego-centric vision, such as Meta's EgoMAN (https://arxiv.org/pdf/2512.16907), show that spatial object features heavily dictate human movement intent but require massive computational resources unsuitable for real-time networking.
This internship will focus on extracting lightweight spatial/semantic context (e.g., 3D object positions and goal descriptors) from ego-centric vision into compact 3D coordinate/vector streams and integrating them into low-latency motion predictive pipelines for hand-arm teleopeartion. The emphasis is on methods that ar computationally feasible for edge deployment and can be evaluated in realistic teleoperation tasks to enable real-time, robust teleoperation over long distances.
Key Research Areas:
- Lightweight Spatial Abstraction: Developing methods to extract critical 3D spatial information and object coordinates $(x, y, z)$ from ego-centric vision pipelines to minimize communication overhead.
- Context-Aware Motion Prediction: Integrating lightweight spatial/object context into time-series predictive models (such as RNNs or TCNs) for hand and arm trajectories, to enhance prediction accuracy under severe network delay and jitter. This includes comparing “environment-blind” baselines against context-conditioned models and exploring prediction horizons relevant for teleoperation.
- Edge-friendly Integration for Teleoperation: Designing and prototyping a low-overhead, end-to-end real-time prediction pipeline that fits within realistic latency requirements while maintaining spatial/semantic awareness.
Preferred Intern Educational Level
Currently pursuing a PhD or a Master’s degree with a strong interest and background in research.
Skill sets or Qualities
- Proficiency in C++ and Python (system development, data analysis, and visualization)
- Solid understanding of Computer Networking and/or 3D Computer Vision/Graphics.
- Proven ability to formulate research questions, solve complex problems, and communicate findings through academic writing.