My research focuses on:
- Real-Time Systems (RTS): task scheduling models, timing analysis and verification, feedback scheduling, real-time operating systems (RTOS), real-time programming languages, control and real-time scheduling co-design and real-time communication.
- Robotics and Autonomous Systems (RAS): vision-based localization and tracking, computation efficiency, sensor insufficiency, high performance architecture, deep reinforcement learning (DRL) and transfer learning (TL).
- Cyber-Physical Systems (CPS): Model-based systems engineering (MBSE) in CPS, model-based safety assurance of CPS, autonomous driving, intelligent transportation / smart city.
Below is a list of reseach projects I have been involved as a RA/PhD:
(2019 - 2022) MOCHA: Modelling and Optimising Complex Heterogenous Architectures
PostDoc, Huawei Funded, 1.2M€
The applications and resources (processors, networks and memory) for real-time systems are becoming ever more complex to understand, control and maintain. This has led to research into building statistical models of systems and adaptive policies based on these statistical models.
The key challenges that emerge are whether the models reflect how the system would behave during operation, how systems should deal with unexpected or rarely occurring scenarios, and then how to optimise systems based on the behaviours of the systems. It is specifically to address the high overheads of current systems and the low cache hit rates that are currently achieved.
(2018 - 2019) DEIS: Dependability Engineering Innovation for CPS
PostDoc, EU-funded Horizon 2020 project, 4.9M€ [homepage]
Cyber-Physical-Systems (CPS) provide the potential for vast economic and societal impact in domains such as automotive, health care and home automation. The open and cooperative nature of CPS poses a significant new challenge in assuring dependability. The DEIS project addresses this important and unsolved challenge by developing technologies that enable a science of dependable system integration. Such technologies facilitate the efficient synthesis of components and systems based on their dependability information. The key innovation in the approach of the DEIS project is the concept of Digital Dependability Identity (DDI). A DDI contains all the information that uniquely describes the dependability characteristics of a CPS component. DDIs are used for the integration of components into systems during development as well as for the dynamic integration of systems into systems of systems in the field.
(2015 - 2018) ATAS: Adaptive Task Scheduling Framework for CPS
PhD research project, University of York
In a Cyber-Physical Control System (CPCS), there is often a hybrid of hard real-time tasks which have stringent timing requirements and soft real-time tasks that are computationally intensive. The task scheduling of such systems is challenging and requires flexible schemes that can meet the timing requirements without being over-conservative.
Fixed-priority scheduling (FPS) is a scheduling policy that has been widely used in industry. However, as an open-loop scheduler, FPS has low system dynamics and no feedback from historic operation. As the working conditions of a CPCS will change due to both internal and external factors, an improved scheduling scheme is required which can adapt to changes without a costly system redesign.
In recent years, there is a large research interest in the co-design of control and scheduling systems that explicitly considers task scheduling during the design of a controller. Many of these works reveal the possibility of adapting control periods at run-time in order to accommodate varying resource requirements and to optimise CPU utilization. It is also shown that control quality can be traded off for resource usages.
In this study, an adaptive real-time scheduling framework for CPCS is presented. The adaptive scheduler has a hierarchical structure and it is built on top of a traditional FPS scheduler. The idea of dynamic worst-case execution time is introduced and its cause and methods to identify the existence of a trend are discussed. An adaptation method that uses monitored statistical information to update control task periods is then introduced. Finally, this method is extended by proposing a dual-period model that can switch between multiple operational modes at run-time. The proposed framework can be potentially extended in many aspects and some of these are discussed in the future work. All proposals of this thesis are supported by extensive analysis and evaluations.
I also actively cooperative with researchers from other projects. Below are some projects that I am/was involved as a collaborator:
(2019 - 2020) CyPhyAssure [homepage]
- contributed to formal methods for safety cases using SACM.
- contributed to tool integration of safety case and Issabelle server.
(2019) FiC: Future factories in the Cloud [homepage]
- Proposed using timing-related statisitcs to form dynamic geo-fencing systems.
- contributed to timing analysis methods of geo-fencing systems.
(2019) AAIP: The Assuring Autonomy International Programme [homepage]
- contributed to failure mitigation of sensory data.
- involved in a various of discussions/meetings.