Researchers at the University of Texas (Austin) and the U.S. Army Research Laboratory are using interactions with a human instructor to show robots or computer programs how to improve their performance. The human provides real-time feedback which has helped create a new algorithm called Deep TAMER, or Training an Agent Manually via Evaluative Reinforcement.
These machine learning algorithms are loosely inspired by the brain in providing a way to learn how to perform tasks by viewing condensed videos streaming in short amounts of time. Many current techniques in artificial intelligence require robots to interact with their environment for extended periods of time to learn how to perform a task. The benefit of these algorithms is that providing feedback on negative operations can be reinforced and “learned” more quickly.
Within the next two years, researchers feel this approach could aid in improving robotic performance in several ways, including autonomous machines that work side-by-side with soldiers on the battlefield. Applications could include driverless vehicles and ordinance disposal. The ultimate goal will be machines that can quickly and safely learn from their human counterparts in styles that range from demonstration and natural language instruction to constructive critiques.