In this project we address the problem of training models and agents to perform various human tasks. The literature dealing with this field offers different approaches that deal with the training problem in diverse ways. All the methods are based on a combination of recognized principles in machine learning such as RL, BC, and original strategies for improving the quality of learning and adapting the algorithms to the nature of the task of imitating human movement.
The method we propose to deal with the imitation problem is based on two central principles. One, which is repeated in a large part of the existing approaches in the literature, is the division of the action to be imitated into a collection of sub-actions, where the sub-actions are used as building blocks to perform the complex action. The second principle is the selection of the sub-operations at each decision point through a combination of principles in image processing and classification networks.
We examine the quality of the results of the method we used against existing methods while comparing the success rate on the same test set.