AI specialists from the US Army laboratory and developers at the University of Texas have created the Deep TAMER artificial intelligence algorithm that can learn new skills faster by watching videos at an accelerated rate. The person who trains the AI will observe the learning outcomes, praise or scold the robot depending on the situation. The learning process itself is reminiscent of dog training, and the speed with which the AI will learn new "tricks" should grow exponentially.
The study's lead author and technology developer Dr. Garrett Warnell explained that with the new AI training program, robots will be able to learn much faster, adapting to circumstances and adapting to current tasks. In addition, he noted that this work will be an important step in the creation of fully autonomous devices capable of independently making decisions and performing a number of complex tasks not only in laboratories, but also in the field.
Now robots learn for a long time, interacting with objects by trial and error. Only after a series of incorrect sequences of actions do they begin to act more or less correctly, but during training, the robot, for example, can completely break down, falling off a cliff or drowning. The new method and help from people will speed up learning and help avoid fatal mistakes.
Viacheslav Larionov