Publications
Localisation-Aware Fine-Tuning for Realistic PointGoal Navigation
Accepted to Towards Autonomous Robotic Systems Conference (TAROS 2024)
Accepted to Towards Autonomous Robotic Systems Conference (TAROS 2024)
Prior research demonstrated the effectiveness of end-to-end reinforcement learning for PointGoal navigation tasks within indoor environments. Given 2.5 billion frames of experience, a navigation policy can be trained to achieve a success rate of 0.94 when deployed in unseen environments.
A limitation of this approach is its reliance on perfect localisation, which is unrealistic for real-world deployment scenarios where localisation must be estimated, inevitably introducing errors.
This work studies the effectiveness of integrating a traditional vision-based SLAM algorithm with a reinforcement learning-based PointGoal navigation policy.
We demonstrate how fine-tuning a pretrained navigation policy on realistic localisation estimates can increase the success rate by 14% (0.71 → 0.85) and SPL by 15% (0.66 → 0.81) when compared to deploying policies in a zero-shot manner.
Machine Fault Detection in Manufacturing using Deep Learning Methods
Accepted to World Automation Conference (WAC 2024)
Accepted to World Automation Conference (WAC 2024)
This project applied deep learning methods to the problem of diagnosing faults within electrical machinery.
Used a transfer learning approach to detect rotor and bearing faults within an induction motor.
A pretrained ResNet-34 was used to extract features from infrared images, and vibration signals were encoded using the Gramian Angular Fields algorithm.
The best performing model achieved a mean test accuracy of 98%.
Projects
Teaching
Robotics & Image Processing - Lead Lecturer (2022/23)
Introduction to C++ - Teaching Assistant (2023/24)
Engineering Optimisation with Python - Teaching Assistant (2023/24)
