|
Abstract:
|
The need of designing and implementing software to prepare robots for the execution of
new tasks implies an expensive cost that requires specific hardware, software and knowledge.
Learning by demonstration is a paradigm for enabling robots to autonomously
perform new tasks learning from previous demonstrations. This project focuses on how
to facilitate the learning of new tasks by robots through demonstrations performed by humans.
In order to accomplish this goal, this thesis considers two subproblems: imitation
and correspondence. Three different algorithms are used in order to solve both subproblems,
in addition to a reinforcement learning to improve the final solution through new
demonstrations. Moreover, a methodology is proposed to perform experimentation using
these algorithms, including a final discussion over their performance and future work. |
