Edge/Fog Computing System Area

  • Edge Computing is a distributed computing paradigm that brings computation and data storage closer to the sources of data
  • Fog Computing is an architecture that uses edge devices to carry out a substantial amount of computation, storage and communication locally and routed over the Internet backbone
  • The DMS Edge/Fog System team is currently researching about
    • Autonomic Provisioning Edge Cluster System
    • Smart Gateway for onboarding IoT devices to Edge Cluster

Cloud Computing System Area

  • Cloud computing is the delivery of computing resources as a service.
  • Cloud computing can be rapidly provisioned and released with minimal management effort or service provider interaction.
  • Cloud Computing System team is currently researching about
    • Autonomic Provisioning Cloud Computing System for Customization
    • Serverless computing
    • Reinforcementt Learning for Cloud Computing

Mobile Robot Navigation Area

  • Global Navigation
    • Global Navigation involves planning the optimal trajectory for mobile robots (both ground and aerial) to move from a starting point to a destination based on a given map.
    • In known environments, algorithms such as A*, Dijkstra, and RRT are used for path planning, while in unknown environments, Simultaneous Localization and Mapping (SLAM) enables real-time map generation and updates.
    • Our research extends beyond wheeled robots (e.g., TurtleBot) to include Aerial robots (e.g., drones), incorporating 3D SLAM and path optimization techniques for aerial navigation.
  • Local Navigation
    • Local Navigation ensures that an agent follows the planned trajectory while dynamically avoiding obstacles and adapting to real-time environmental changes.
    • Avoiding not only static obstacles but also dynamic obstacles such as humans, vehicles, and drones is a key challenge.
    • We leverage Deep Reinforcement Learning (DRL) and Multi-Agent Reinforcement Learning (MARL) to enable robots to learn and make optimal navigation decisions in complex, dynamic environments.
    • MARL-based approaches allow multiple agents (both ground and aerial robots) to cooperate in real time, minimizing interference and optimizing collective movement strategies.
    • Our research focuses on both wheeled mobile robots (TurtleBot, etc.) and aerial mobile robots (UAVs etc.), ensuring efficient and safe navigation in 3D environments with dynamic obstacles.
  • Digital Twin
    • Digital Twin technology creates realistic digital replicas of real-world robotic systems, providing a high-fidelity simulation platform for testing and optimizing algorithms before real-world deployment.
    • Using Unreal Engine, we develop high-resolution simulation environments, allowing us to train and validate reinforcement learning-based navigation models efficiently.
    • Our research includes building Digital Twins for both UAVs and ground robots, enabling predeployment training and validation of reinforcement learning models in diverse scenarios.
    • We specifically focus on Multi-Agent Digital Twin Simulations, where multiple mobile robots operate in a shared virtual space, enabling collaborative navigation, obstacle avoidance, and optimal coordination strategies.

Computer Vision AI Area

  • Computer Vision is one of the latest areas of computer science that studies the part of a machine’s vision
  • The DMS Vision team is working on an interactive healthcare cleaning robot system for the Silver Generation.
    • Object detection to detect people
    • Face recognition to recognize a particular user
    • Emotion Recognition for monitoring user emotions
    • Action Recognition to recognize the user’s motion
    • Pose Estimation to perform the user’s exercise assistance role

Natural Language Processing AI Area

  • Natural Language Processing (NLP) is a field of AI that focuses on the interaction between computers and humans using natural language.
  • The NLP AI team is currently researching about:
    • LLM Agent prompt optimization techniques
    • Parameter-Efficient Fine-Tuning (PEFT) methods for large language models
    • Development of LLM Agent for Silvercare Assistant
  • Our team focuses on:
    • Developing prompt optimization techniques using Small LMs as prompt generators
    • Exploring Direct Preference Optimization (DPO) for efficient fine-tuning
    • Implementing Low-Rank Adaptation (LoRA) for efficient model adaptation
    • Designing multi-agent systems for complex NLP tasks
    • Enhancing memory capabilities of LLM Agents for improved performance

Digital Twin Area

  • Digital twin (DT) is a pioneering technology and a promising game-changer in various emerging industries.
  • The concept of DT opens a world of possibilities for fundamentally the infusion of physical-twin specific computational models which are dynamically updated into a feedback loop of data-driven analysis and decision-making.
  • A digital twin is a set of coupled computational models that gradually transit throughout different states in its featured state-space as time goes on, in which constantly and equivalently represent the real-world structure, behaviors and surrounding context of its physical twin.
  • DMS Group is working on the development of (i) neural digital twin dynamic engines (DTDE), (ii) neural digital twin control engines (DTCE), (iii) digital twin control frame (DTCF) and (iv) digital twin cloud infrastructure (DTCI) for U*V systems.

Dependability and Security Quantification Area

  • Dependability and security are of five distinctive natures (along with functionality, performance, and cost) for computing and communication systems.
  • Computing systems and networks with a sophisticated composition of multi-level systems and things are inevitably prone to a chain of threats (faults, errors, and failures) which eventually causes fatal losses, such as service interruption/outage, data leak, or even human lives.
  • Even a 1% failure rate is too high, because it causes 3.65 days of unscheduled downtime in a year which, in turn, may reduce an enormous amount of enterprise turnover.
  • Therefore, dependability and security requirements should be taken into consideration to obtain the highest level of trustworthiness for computing infrastructures, in practice.
  • DMS Group is working on the quantification methodologies for dependability and security metrics of computing systems and networks: virtualized server systems (VSS), data center networks (DCN), software defined network (SDN), Cloud-Fog-Edge Continuum (CFE), Internet of Medical Things (IoMT), Internet of Industrial Things (IoIT), unmanned aerial systems (UAS) etc.