Research Summary

My main research interests and contributions lie within the areas of control systems, artificial intelligence, mechatronics, robotics, guidance, control and automation of unmanned ground and aerial vehicles. The main motivation of my research has always been to inject “learning” phenomena into various types of control algorithms as well as their real time implementations.

My contributions to interval type-2 fuzzy neural networks have made a notable impact on the computational intelligence community and have also received significant international recognition. The outcomes and findings of my research have been regularly peer reviewed and accepted for the best journals in the relevant field – IEEE Transactions and premier international conferences, such as Fuzzy Systems (FUZZ-IEEE).

I am the first author of a course book “Fuzzy Neural Networks for Real Time Control Applications, 1st Edition Concepts, Modeling and Algorithms for Fast Learning”, published by Elsevier. Among the other books in type-2 fuzzy logic theory, I believe that this book has the most comprehensive stability analysis for type-2 fuzzy logic systems.

During my post-doctoral research at the University of Leuven (KU Leuven), my field of expertise was the real time implementation of unmanned ground robots. I focused on system modeling, identification and control of large scale systems, complex mechatronic systems and robotics guidance. Within the framework of agricultural robotics, my group and I designed a fully autonomous tractor-implement system, and equipped it with various sensors and actuators.  The outcome of this project was a practical mechatronic system, illustrating how control, sensing, and actuation can be integrated to achieve an intelligent system.

I am currently investigating guidance and vision-based control of unmanned aerial robots. Visual tracking is a critical task in numerous applications and has attracted a lot of attention in the computer vision community. However, an ideal tracker that outperforms others in all situations does not exist due to the difficulties of recognizing a target undergoing occlusion, rotation, appearance variations or illumination changes. Currently, I have two ongoing industrial research projects. The first focuses on the guidance of unmanned aerial robots by using novel tracking algorithms. The second project aims to implement learning control strategies on unmanned aerial vehicles to improve their control accuracy as well as their capabilities to function under uncertain working environments.

Not only drones but also all different applications of robots are going to be one of the most significant technologies for the next few decades. Therefore, in the future, I hope to focus on unmanned ground and aerial vehicles and their daily life applications by enriching their capabilities using vision-based algorithms. The future of drones is very promising and will be an area of ongoing development. Therefore, my long term research agenda includes creating a link between control theory (in particular learning control strategies), artificial intelligence, robotics and vision.

My Youtube channel :

https://www.youtube.com/channel/UCZW3-zCmwIoZWTW3Gb6PwJw

Interests

  • Artificial intelligence
  • Intelligent control
  • Robotics
  • Mechatronics
  • Sliding mode control
  • Model predictive control
  • Precise landing for unmanned aerial vehicles (by ST Eng-NTU Corp Lab, $470,000) (2015-2017)

    Precise landing for unmanned aerial vehicles (by ST Eng-NTU Corp Lab, $470,000) (2015-2017)

    ST Eng-NTU Corp Lab

    Without loss of generality, there exist two basic methods for the realization of UAV automatic guidance: local positioning systems (vision or laser based sensors) and global positioning systems (GPSs). Today’s fast moving technology allows us the application of real-time kinematic (RTK)-global positioning systems (GPSs) which can provide an accurate positioning accuracy of a few centimetres. However, they may not be very accurate for a precise landing. As an alternative, vision-based systems are cheap to implement. However, they have some disadvantages in outdoor environments, e.g. to be very sensitive to light conditions. In this project, the advantages of local and global positioning systems will be combined to realize one specific goal: precise landing. The UAV will use the GPS until a certain border which will be an artificial sphere around the target point. After the artificial sphere, the navigation method will be stitched from GPS to a low cost camera.

  • Fuzzy neural network-based learning control of unmanned aerial vehicles (by ST Eng-NTU Corp Lab, $130,000) (2016-2017)

    Fuzzy neural network-based learning control of unmanned aerial vehicles (by ST Eng-NTU Corp Lab, $130,000) (2016-2017)

    ST Eng-NTU Corp Lab

    The performance of a model-based controller is quite satisfactory in case of having a precise mathematical model of the system as well as when uncertainties and parameter changes are negligible. However, for the case of small size UAVs, it is a difficult, time consuming and expensive task to obtain the precise model of the system. Moreover, such small UAVs are more vulnerable to transient environmental conditions which will destabilize them.

    In such cases, a model-based controller with a learning capability or a model-free controller is preferable; both methods will be implemented throughout this project. For the online learning purpose in the former method, the combination of artificial neural networks and fuzzy logic controllers will be implemented for the control of small size UAVs. In the latter method, artificial neural networks or fuzzy neural networks will be preferred. The most significant advantage of learning capability is that it is able to expand the operating envelope that cannot be modelled precisely in real-life. Thus, the system dynamics can be learned on-line.

  • Automated Construction Quality Assessment Robot System (A-CONQUARS) (by JTC Corporation - NRF, $821,160) (2015-2016)

    Automated Construction Quality Assessment Robot System (A-CONQUARS) (by JTC Corporation - NRF, $821,160) (2015-2016)

    Collaborators: Prof Chen I-Ming, A/P Robert Tiong, NTU, CTRLWORKS Corp.

    New buildings in Singapore may soon have a high-tech building inspector rolling up to their door steps armed with laser scanners and high-tech cameras that can spot the tiniest cracks and defects.

    This new building inspector is a robot invented by scientists from Nanyang Technological University, Singapore (NTU Singapore), co-developed with Singapore’s national industrial developer JTC and local start-up CtrlWorks.

    Named QuicaBot – short for Quality Inspection and Assessment Robot – it can move autonomously to scan a room in minutes, using high-tech cameras and laser scanners to pick up building defects like cracks and uneven surfaces.

    A few of these robots working together will make inspecting a building a breeze. The robots can upload 3D data of the scans to the cloud and inform the human operator, who can then inspect critical and complex defects.

    “Visual inspection of a new building is an intensive effort that takes two inspectors, so we have designed a robot to assist a human inspector to do his job in about half the time, saving precious time and manpower, and with great accuracy and consistency,” explained Prof Kayacan.

  • Model predictive control-moving horizon estimation framework as applied to tilt rotor UAVs and its experimental evaluation (by Nanyang Technological University, $88,000) (2015-2017)

    Model predictive control-moving horizon estimation framework as applied to tilt rotor UAVs and its experimental evaluation (by Nanyang Technological University, $88,000) (2015-2017)

    Tier 1 Project

    The main goal of this project is to design model predictive control (MPC)-moving horizon estimation (MHE) framework for highly nonlinear unmanned aerial vehicles (UAVs). In order to reach the main goal, the following sub tasks will be realized in real time:

    1. Design and realization of a twin rotor system to solve the manoeuvrability problem,
    2. Design and realization of a tilting rotor system to solve the stability and transition problems,
    3. Design and realization of a tricopter system to solve the take-off problem,
    4. Design and realization of a tilt rotor UAV as a final test bend,
    5. Performance evaluation of MPC-MHE framework for the control of highly nonlinear UAVs (a tilt rotor UAV in this project).
  • Learning control algorithms for unmanned aerial vehicles. (by Nanyang Technological University, $100,000) (2014-2017)

    Learning control algorithms for unmanned aerial vehicles. (by Nanyang Technological University, $100,000) (2014-2017)

    Start Up Grant

    The performance of a model-based controller (e.g. a model predictive controller (MPC)) is quite satisfactory in case of having a precise mathematical model of the system as well as uncertainties and parameter changes are negligible. However, in practical applications, it is a challenging task to determine an accurate model of a mechatronic system. Moreover, the system to be controlled is always subjected to both internal and external noise, and it has to operate under uncertain working environments. In such cases, a model-based controller with a learning capability or a model-free controller is preferable.

    The research interest in the control and navigation of unmanned aerial vehicles (UAVs) has significantly increased in recent years. This trend seems to continue growing in the near future since the application areas for UAV are wide spread from remote delivery of urgent materials to reconnaissance purposes including all military and civilian applications.     Among all the required endeavors in software and hardware of a UAV, the most prominent goal is the completion of a mission completely autonomously. In order to be able to take off, do a specific pre-defined mission and land on successfully, the UAV must have sophisticated estimation and control algorithms. Since it is difficult to obtain precise models for UAVs and they have to operate under uncertain working conditions, the main goal of this project is to design several learning model-based and model-free control techniques for the control of UAVs. As a model-based approach, linear and nonlinear model predictive controllers will be elaborated with some estimation/learning techniques, such as extended Kalman filters (EKFs) and moving horizon estimators (MHEs). As a model-free method, the combination of neural networks and fuzzy logic controllers will be studied.

  • Design of lightweight UAV for 3D Printing (by NRF, $2,052,500) (2014-2024)

    Design of lightweight UAV for 3D Printing (by NRF, $2,052,500) (2014-2024)

    NRF Medium-Sized Centre (MSC)

    Carbon composite structures are the most preferred materials that are used to manufacture UAV structures. Their high strength to weight ratio is an advantage but at the same time, they are costly. Besides, the cost of manufacturing complex geometries would require moulds which would again add into the cost and time for the fabrication of these parts. Critical design details such as an aerodynamic twist might not be accurately realized in this method of fabrication. In addition, to produce robust carbon fibre parts, the mechanics is complex and failure can occur frequently. Furthermore, the repair or replacement of broken carbon fibre part is again another costly affair. In contrast, in order to manufacture complex UAV structures, 3D printing method would be an advantage. 3D printing does not require any moulds or fixture which would drastically reduce the cost and time. Adding details to the complex structures during the manufacturing process would be time consuming and labour demanding. However, modification of the 3D printed part would be easier since adjustments would be done directly on the CAD software.

    Making use of the advantage of 3D printing, the design and manufacturing of the VTOL UAV is proposed in this project. Later on, the following phases of this project proposal explore the novelty of the VTOL transition flight, mission flight profile (hovering, transition, cruise and landing) modelling and flight controller design. Finally, flight trials of the 3D printed VTOL UAV will be conducted on the final phase of the project.

A few videos from my research at NTU, Singapore

Quicabot: Quality Inspection and Assessment Robot - Reuters

Quicabot: Quality Inspection and Assessment Robot

Quicabot: Quality Inspection and Assessment Robot

This UAV helps you to find the emergency exit in case of an emergency situation

RRT-based 3D Path Planning for Formation Landing of Quadrotor UAVs

Autonomous Navigation of UAV by Using Real-Time Model-Based Reinforcement Learning

Vision-based Autonomous Tracking and Landing on a UGV: Project Demo

Type-2 Fuzzy Logic Control of UAVs

Input Uncertainty Sensitivity Enhanced Non-Singleton FLC for Quadrotor UAVs

Single Input Type-2 Fuzzy PID Control of UAVs in Real Time

Single Input Type-2 Fuzzy PID Control of 3 DOF Helicopter Testbed (WCCI 2016, Vancouver, Canada)

Tracking Recommendation Detection

Vision-based Autonomous Tracking and Landing on a UGV - Flight During Ho Ching's Visit

Precise landing for unmanned aerial vehicles - indoor and outdoor test

Precise landing for unmanned aerial vehicles - hovering test

EID Project at NTU

Object Detection Algorithms for Boat Detection

Ultra Lite Launcher

NTU MAE FMC Lab 2014-2015 FYP and URECA Projects

A few videos from my research at KU Leuven, Belgium

Autonomous tractor at KU Leuven

Autonomous tractor at KU Leuven