Learning from Demonstration with Gaussian Process Approach for a Omni-directional Mobile Robot

作者：Sillas D. G.^*; Hurtado E. G.; Vargas E. S.; Rivero C. R.; Delgado G. D.

来源：IEEE Latin America Transactions, 2018, 16(4): 1250-1255.

• gaussian process
• machine learning
• regression
• supervised learning
• robotics

There is great potential in robotics for the application of methodologies related to machine learning, and aiming to the learning from demonstration framework, which helps in a remarkable way to reduce the programming times of the movements of a Robot and in addition to this allows the movements to be more natural and smooth. Machine learning can be found in diverse derivations as the supervised, nonsupervised and reinforced learning. One of the most used is supervised learning, which can be divided into two main tasks: classification and regression. To perform regression, one of the methodologies widely used is the Gaussian Process (GP). Learning from Demonstration (LID) can be seen as a regression problem given a data set. In this article, the development of a learning method based on Gaussian Process and applied to a mobile robotic platform is presented. This implies that the learning has to be performed very fast and incrementally, this because the data arrives frequently and abundantly. The GP offers great precision in learning and meets the characteristics to perform control in robots. This paper investigates the regression methodology using GP as well as its application in demonstration learning.

• 出版日期2018-4