Repository@USM

Welcome to closed access digital repository of Universiti Sains Malaysia (USM).
This repository contains multiple types of scholarly materials, especially USM theses and exam papers.

“Any doctoral thesis or academic work owned by Universiti Sains Malaysia (USM) that is downloaded through its official platforms shall not be shared, distributed, printed, or reproduced in any form to third parties without explicit written permission from USM. Any breach of this condition constitutes a violation of institutional intellectual property rights and may result in disciplinary action, legal proceedings, or revocation of academic credentials, depending on the severity of the offence.”

To access the full text, please log in with your USM email account.

Photo by @clarissemeyer

Research outputs

7998

Projects

People

Recent Additions

Reinforcement learning-integrated hierarchical multi-agent Framework for cooperative indoor exploration.
(2026-10)
Low, Derrick
Autonomous exploration of unknown indoor corridor networks by nano-quadrotors is challenging because achieving verifiable complete branch coverage requires no metric map and no external localisation, under the sensing limits of a four-directional Time-of-Flight and optical-flow sensor suite matched to the Crazyflie CF2.X platform. Existing frontier-based methods require an occupancy grid that degrades under odometric drift, while micromouse algorithms provide complete coverage guarantees but assume discrete grid representations and binary wall sensors unavailable on continuous-environment platforms. A further gap exists at the multi-agent level: when multiple agents explore the same corridor network cooperatively, peer body artefacts corrupt junction classification, and standard branch-memory approaches do not partition exploration work or prevent concurrent branch entry without a shared metric map.This thesis addresses both gaps with a four-layer hierarchical architecture that combines a learned, width-invariant junction classifier with a generation-based topological memory and a deterministic rule-based coordination layer, extended to cooperative multi-agent operation without joint training or centralised control, and evaluates both single-agent and multi-agent configurations in simulation. The system was implemented in PyBullet using hardware-matched Crazyflie CF2.X parameters; all results are obtained in simulation, with hardware parameter matching intended to support, not substitute for, future physical validation. A Proximal Policy Optimisation classifier was trained via a four-phase curriculum to identify fourteen discrete junction types from a twelve-dimensional observation vector of width-normalised range readings and onboard heading, with no position information supplied. A generation-based topological memory records per-branch exploration status for every confirmed junction,and a depth-first search coordinator provides a formal coverage guarantee that every reachable branch is entered exactly once in the forward direction. The multi-agent extension requires no modification to the learned junction classifier: it adds a branch status ownership protocol via a shared junction database, an explorer and supporter role-assignment framework, and peer body masking, enabling cooperative exploration as a Decentralised Partially Observable Markov Decision Process with Communication. Four independent out-of-distribution sweeps spanning corridor width, branch length, and heading noise gave 77.4% pooled accuracy, with the sharpest degradation driven by branches shorter than the corridor is wide rather than by corridor width itself. The single-agent system achieved 100% coverage completeness across three maze grid sizes and a 2× corridor width range (0.6–1.2 m) without retraining. Multi-agent exper iments scaling from one to four agents achieved at least 99% mean coverage at every agent count, with scaling efficiency falling from 94–97% at two agents to 65.7–72.1% at four agents as coordination overhead grows relative to the shrinking exploration territory assigned to each agent, confirming that the branch-status protocol achieves complete work partitioning without joint training or centralised coordination even as agent count increases. These results indicate that topological memory with position proximity matching, paired with a minimally learned perception module, is a practical route to infrastructure-free indoor exploration on sensing-constrained nano-quadrotor platforms, with direct application to search-and-rescue and infrastructure-inspection settings where metric maps and external localisation are unavailable.
Metaheuristic-driven control optimization for hybrid quadplane UAV in multimodal flight operations.
(2026-08)
Wang, Shirui
Hybrid quadplane unmanned aerial vehicles (UAVs), which combine the vertical takeoff and landing capability of rotary-wing platforms with the long-range and high-efficiency cruise performance of fixed-wing aircraft, have attracted increasing attention for complex and endurance-critical missions. However, reliable autonomous operation across rotary-wing, fixed-wing, and transition flight regimes remains challenging. Existing studies still exhibit limitations in three main aspects: multimodal dynamic modeling often struggles to simultaneously achieve high modeling accuracy and computational efficiency; transition control strategies mainly depend on direct controller switching or simplified allocation methods, which may cause discontinuous responses; and metaheuristic-based controller optimization remains vulnerable to premature convergence and insufficient population diversity. To address these challenges, this thesis proposes a unified framework combining multimodal dynamic modeling,mode-dependent control strategy design, and enhanced metaheuristic-driven controller optimization for hybrid quadplane UAVs. First, flight-test-based multimodal dynamic models are developed for rotary-wing, transition, and fixed-wing regimes using system identification techniques. The proposed approach provides experimentally validated dynamic models with improved accuracy and practical computational efficiency. Based on these models, dedicated controllers are designed for different flight modes, including cascaded PID control for rotary-wing flight, PID-TECS control for fixed-wing flight, and an airspeed-based control-weight allocation strategy for transition flight. Simulation results demonstrate that the identified models achieve model–data matching rates exceeding 84% across all flight regimes, while the proposed transition strategy effectively eliminates the abrupt attitude change caused by mode switching. Second,an enhanced metaheuristic optimization algorithm, termed the Eagle Hawk–Artificial Rabbit Optimization algorithm, is proposed to improve controller parameter tuning. EHARO introduces an Eagle Hawk Threat-Based Perturbation Mechanism and a self centric exploration strategy to enhance population diversity and global search capability. Evaluations using 29 benchmark functions from the CEC2017 test suite demonstrate that EHARO achieves superior optimization performance compared with several state of-the-art algorithms. When applied to hybrid quadplane attitude control, EHARO optimized PID controllers achieve faster dynamic responses and improved attitude stabilization across roll, pitch, and yaw channels. In particular, the roll-channel settling time is reduced by more than 8.2%, while reductions exceeding 52% and 75% are achieved in the yaw channel compared with IARO and ARO, respectively. Overall, this research establishes a unified modeling, control, and optimization framework that provides an effective approach for improving modeling accuracy, transition stability, and controller performance in hybrid quadplane UAVs operating in multimodal flight environments.
Design and development of an S-Band Continuous wave radar prototype Capable of Enhanced target Differentiation.
(2026-07)
Zulkarnain, Siti Aminah
The demand for a target detection system that are capable of identifying objects is continuously increasing in the fields of security, surveillance, and aviation. This demand is particularly relevant when differentiating between micro unmanned aerial vehicles (MUAVs), birds, human motions, and others. Conventional radar systems have difficulty differentiating targets especially with comparable radar cross-sections (RCS) and overlapping motion characteristics that result in false alarms. This research addresses these challenges by designing an S-band continuous wave (CW) radar prototype capable of enhanced target differentiation through motion analysis using micro-Doppler (M-D) features. The radar system was first designed and simulated in low level processing technique. Then, the radar system was developed using Commercial Off-the-Shelf (COTS) components that operate at a carrier frequency of 3.4 GHz. The Short-Time Fourier Transform (STFT) was used to analyze motion data in order to extract M-D features. Simulation studies revealed the capability to differentiate between rotating MUAV blades ranging in length from 20 to 40 cm and flapping bird wings measuring 35 cm in total span based on their distinct modulation effects in the frequency domain. The performance of the developed prototype was assessed experimentally to differentiate between three types of human movement defined by average pace, namely slow walking at approximately 2 m/s, fast walking at 5 m/s, and running at 7 m/s. Each of these movements displayed increasing Doppler spreads in the STFT spectrogram. MUAV tests were conducted using a DJI Mini 3 under two common scenarios, like hovering and flying motion. Hovering motion produced clear periodic M-D patterns due to blade rotation, while flying motion in a translational direction resulted in five distinct Doppler bursts with superimposed ripple patterns. This suggests that the simultaneous detection of both translational velocity and propeller blade motion. The results demonstrate that the designed S-band CW radar system provides reliable detection and motion analysis across diverse target types.
Efficient obstacle detection and approach control for Micro air Vehicles using optical flow and texture cues.
(2026-09)
Shi, Xu
Micro air vehicles (MAVs) operating in confined and GPS-denied environments require lightweight visual methods for short-range approach control and obstacle triggered stopping. Optical-flow divergence provides a compact cue for regulating approach motion, but it remains primarily a motion-related signal and does not explicitly determine whether an observed appearance change corresponds to an obstacle. This limitation motivates the integration of an additional appearance-based cue. This thesis proposes a monocular onboard framework that combines constant divergence optical-flow control with texton-based visual appearance analysis. The optical-flow module regulates the MAV forward approach using sparse motion cues, while the texton-based module identifies sustained appearance deviations through a compact texture dictionary, chi-square histogram comparison, and temporal confirmation criterion. Both modules operate in parallel on a monocular smart camera at a resolution of 640 × 480 pixels and an onboard processing rate of approximately 10–15 Hz, without dense environmental reconstruction, deep-learning inference, or external image processing. The framework was evaluated in six indoor flight trials using divergence setpoints of −0.1, −0.2, and −0.3 rad/s. With the obstacle positioned approximately 3.2 m from the forward-flight starting point, the MAV stopped after travelling approximately 2.4–3.0 m, leaving a remaining separation of approximately 0.2–0.8 m. The recorded divergence responses showed approximate approach regulation under the imposed forward-velocity constraint, while the texton-based module generated obstacle-triggered stopping events when the appearance-deviation criterion persisted for the required duration. Spatial likelihood maps generated from representative static images further provided qualitative visualizations of local obstacle-related texture responses. Overall, the results provide preliminary evidence that sparse optical-flow divergence and compact statistical appearance cues can be integrated onboard for short-range MAV approach regulation and obstacle-triggered stopping under the tested indoor, low-speed conditions. The findings demonstrate the feasibility of the proposed integration on a resource-constrained platform, but do not establish general detection reliability or robustness across broader obstacle types, environments, lighting conditions, or flight speeds.
Analysis of low-latitude ionospheric Scintillation variations over Malaysia Using GPS data.
(2026-08)
Alhashimi, Nasser Said Saif
Ionospheric scintillation degrades the accuracy and reliability of GPS signals, particularly in equatorial regions such as Malaysia. This study investigated the temporal and spatial characteristics of ionospheric amplitude scintillation over Malaysia during 2014, the peak phase of Solar Cycle 24, and examined its relationship with solar activity and GNSS signal degradation. GPS observations from UNIMAS, UKM, and LNO were analysed using S4 index. Seasonal, diurnal, and geographic variations were evaluated together with the effects of sunspot numbers, Solar Radio Flux (F10.7), and solar flare events. A machine learning-based approach was applied to reconstruct missing scintillation data using multi-station observations, followed by hidden-data testing and simulation based validation using the Spirent GSS6700 simulator. Results showed that strong scintillation events (S4 ≥ 0.5) mainly occurred during post-sunset hours and in equinoctial months. UNIMAS recorded the highest scintillation intensity due to its proximity to the EIA region. The reconstructed S4 data achieved high prediction accuracy with a regression coefficient exceeding 0.97 while preserving realistic scintillation behaviour. The main contribution of this research is the establishment of an integrated framework that combines solar activity analysis, multi-station scintillation assessment, machine learning-based reconstruction, and GNSS simulation validation within a single study. This research demonstrates the feasibility of combining machine learning methods with GNSS simulations to reconstruct data and evaluate system performance.

Repository@USM

Welcome to closed access digital repository of Universiti Sains Malaysia (USM). This repository contains multiple types of scholarly materials, especially USM theses and exam papers.

To access the full text, please log in with your USM email account.

Welcome to closed access digital repository of Universiti Sains Malaysia (USM).
This repository contains multiple types of scholarly materials, especially USM theses and exam papers.