In a joint effort, University of Texas at Austin and researchers at the U.S. Army Research Laboratory have come up with new techniques for computer programmes or robots so as to learn how to accomplish tasks through interaction with a human instructor. The findings obtained from this research will be presented before and published at the Association for the Advancement of Artificial Intelligence Conference in New Orleans, Louisiana, from February 2 to 7.
The New Algorithm is an Extension of TAMER, an Earlier Version of the Algorithm
Researchers from Army Research Laboratory or ARL and University of Texas have taken into consideration a particular case wherein a human being gives a real-time feedback as a critique. The idea was first incepted, implemented and introduced by collaborator Dr. Peter Stone, professor from the University of Texas in Austin, together with Brad Knox, Dr. Stone’s former doctoral student, as Training an Agent Manually via Evaluative Reinforcement or TAMER. This team of the Army Research Laboratory and University of Texas jointly came up with a new algorithm and named it Deep TAMER.
Deep TAMER is an extension of its earlier version called TAMER that makes use of deep learning. Deep learning refers to class of algorithms that are involved with machine learning and it loosely draws inspiration from the human brain so as to give a robot the capability to learn how to accomplish tasks by watching video streams in a short period of time with a human instructor.
In accordance with Dr. Garrett Warnell, an army researcher, the team has taken into consideration situations wherein a human instructor teaches an agent as to how to behave after making an observation and giving critical views. Dr. Warnell also said that the researchers have extended their earlier work in this very field so as to enable this similar kind of training for robots or computer programs.