Seminar by Ricardo Martins
This work presents the theoretical foundations of the Bayesian models and experimental implementation of the touch attention mechanisms involved in the active haptic exploration of heterogeneous surfaces by robotic hands. The Bayesian model ?tar infers the sequence of regions of the workspace that are explored by the sensory apparatus of the robotic system, taking in consideration the objective of the task, saliency and uncertainty associated to the haptic perception of the workspace and inhibition-of-return mechanisms. The Bayesian model ?per perceives the material category of the surface being explored (haptic stimulus), integrating the haptic sensory inputs from the region of the workspace explored the robotic hand. The Bayesian models proposed in this work were designed to be integrated in robotic systems with different sensory apparatus and mechanical structures.
The experimental results demonstrate that the Bayesian model ?per can be used to discriminate 10 different classes of materials with an average recognition rate higher than 90%. The ATLAS robot, in the simulation, was tested in three scenarios (generalization capability) with different configurations of haptic stimulus. The robotic system has used a robotic finger to perform the search and follow of discontinuities between regions of surfaces of different materials with a divergence smaller than 1cm from the ground truth, using only haptic sensory data (Bayesian model ?tar).