PROJECTS

Learning Continuous LiDAR Navigation in a Randomised Cone Maze

A holonomic robot learns to traverse randomised obstacle fields from a 360-degree LiDAR scan, via expert demonstration, dataset aggregation, and teacher-anchored PPO.

Reinforcement learning
Robotics
ROS 2

2026 Read the write-up → GitHub

Image Inpainting with Deep Learning: Convolutional Autoencoders and Vision Transformers

Designing and training a Convolutional Autoencoder and Vision Transformer for image inpainting.

Computer vision
Deep learning

2026 Read the write-up → GitHub

Localizing Predetermined Objects Underwater Using Stereo Vision

A staged investigation into stereo-based depth estimation for underwater object localization, covering classical disparity methods, deep learning inference, and empirical calibration fitting.