Maliheh Marzani - Computer Science MSc Thesis Defense
Title: Texture Classification on Uneven Surfaces: Using Deep Learning Techniques
Maliheh Marzani
Robots are increasingly vital across various fields, and enhancing them with human-like touch capabilities opens new functional possibilities. Tactile sensors enable robots to perceive and interact with their surroundings similarly to humans. This research focuses on utilizing tactile sensors to classify textures on uneven surfaces, a novel area not extensively covered before. Our methodology involves collecting data points along set paths on an object's surface to create a trajectory for tactile data collection, reducing assumptions about the surface's geometry and enhancing adaptability.
A sliding window approach is used to analyze texture features, segmenting data into smaller, overlapping windows to improve accuracy and reduce computational load. The dataset from uneven surfaces is complemented with data from even surfaces obtained from another study. Advanced deep learning models, including 1D convolutional neural networks (CNNs) and bidirectional long short-term memory (LSTM) networks, are applied for classification. These models achieve high performance, with accuracy, precision, and recall rates of 92.3% for uneven surfaces and higher rates for even surfaces.
This research highlights the importance of integrating tactile sensing into robotics for better texture classification on uneven surfaces. By using bio-inspired sensors like MARG and barometers, the study enhances robotic tactile sensing and addresses the challenge of interacting with varied environments, paving the way for advanced applications requiring precise tactile perception.
Committee members:
Dr. Thiago E. Alves de Oliveira (supervisor)
Dr. Garima Bajwa (Internal examiner)
Dr. Prado da Fonseca (External examiner)
For questions about the seminar, please contact Dr. Todd Randall at dean.ses@lakeheadu.ca.
To register for this event, please email grad.compsci@lakeheadu.ca and a Zoom link will be shared.