Abstract
This thesis addresses the problem of tactile sensing in a robot. We
construct an artificial finger and use machine learning to acquire the
ability to recognise textures and predict slip. The finger has randomly
distributed strain gauges and polyvinylidene fluoride (PVDF) films em-
bedded in silicone.
In the texture recognition task, several machine learning algorithms
such as naive Bayes, decision trees, and naive Bayes trees have been
trained to distinguish materials sensed by the artificial finger. Different
textures induce different intensities of vibrations in the silicone. Conse-
quently, textures can be distinguished by the presence of different fre-
quencies in the signal. The data from the finger are preprocessed and
the Fourier coefficients of the sensor outputs are used to train classifiers.
We show that the learned classifiers generalise well for unseen datasets.
Our classifiers can distinguish between different materials such as carpet,
flooring vinyls, tiles, sponge, wood and polyvinyl-chloride (PVC) woven
mesh with an accuracy of 95 _ 4%.
In the slip prediction task, we predict a slip by studying temporal
patterns in the multidimensional time-series data about the finger-object
contact. The multidimensional time-series is analysed using probabilistic clustering that transforms the data into a sequence of symbols that is used to train a hidden Markov model (HMM) classifier. Experimental results show that the classifier can predict a slip, at least 100ms before the slip takes place, with an accuracy of 96% on unseen datasets.