Discrimination de textures et quantification de rugosité par algorithme d'apprentissage

Abstract

During the last years, the research on robotic object’s manipulation made interesting progresses. In order to improve robot’s dexterity, this essay proposes a new approach to find information about the robot’s neighborhood using tactile sensors.We focus on two main properties : textures discrimination and roughness estimation. We propose three experiments on textures discrimination : the first one discriminates 4 textures with different speeds and strengths between the sensor and the texture during the acquisition. The second one proposes a way to recognize a texture in spite of the initial orientation of the sensor on the texture. The last one discriminates 10 smooth textures showing the sensor’s accuracy. For each experience, the simulations using ANN or SVM and optimized by genetic algorithm have results higher than 90% of recognition. The use of genetic algorithm has an interesting impact and allow to find the best value for different variables. The limit of this kind of algorithm is the necessity of learning each texture you need to recognize. In our second study, we look for a scale to estimate any texture’s roughness based on human roughness feeling. To create this scale we asked 30 people to classify 25 textures chosen from the smoothest to the roughest on a scale from 1 to 10. Then, we create an algorithm to estimate the roughness using these results. We conclude human roughness feeling have difficulties to estimate roughness with precision on a scale from 1 to 10. Despite using 4 different algorithms, we have troubles to generalize the algorithm on unknown textures. A new study with more textures should be necessary.

Date	2 Oct 2014
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Vincent Duchaine (Supervisor)