Tactile Discrimination with Whiskers

Rodents are excellent at tactile discrimination. They use their whiskers to extract tactile infor- mation about the world. Though a huge amount of research has looked at the rodent whisker sys- tem is great detail, it remains unclear what kinds of tactile information can be encoded through a whisker in principle, and how such processing can be performed in practice.
The goal of this thesis is to determine what sort of tactile information can be sensed with a whisker, and to evaluate different methodologies for processing this tactile information. Tactile data is generated using a biomimetic artificial whisker and an XY positioning robot, providing a comprehensive test bed for numerous classifiers of different tactile features. A new framework for whisker based tactile sensing is described, to organise and simplify the task of gathering information. Each part of the framework is then addressed to build a complete system for deter- mining the nature of a surface encountered during exploration.
Previously proposed model-based methods, and new data-driven model-free approaches are compared rigorously on a common data set to establish a benchmark for whisker based tactile sensing. These methods are compared in the task of determining the radial distance to contact of an object, and the texture of a surface under different whisker movement conditions. Classifier robustness is evaluated by testing performance in conditions where different contact parameters are changed simultaneously. The best classifiers in each instance are demonstrated on a se- ries of whiskered robots, generating tactile ‘reports’ that could be used for object identification, localisation and navigation.