Teaching a robot how to walk and jump is one thing, but how about a bot with a knack for dragging its backside across the floor or sliding forward on its nose.
These rather unusual behaviors emerged during Google’s testing of a new technique to allow simulated robots to independently learn useful skills, such as walking and jumping, which can be applied to a wide variety of tasks.
“Conveniently, because skills are learned without a priori knowledge of the task, the learned skills can be used for many different tasks,” the researchers note in the paper.
SEE: IT leader’s guide to the blockchain (Tech Pro Research)
The researchers from the Google Brain team and UC Berkeley say their new approach to unsupervised learning can lead to novel and amusing behaviors, such as the simulated cheetah scooting forward on its nose, sliding forward on its backside or running backwards.
“This observation motivates our claim that human designers simply do not know or cannot imagine some skills agents can acquire,” the paper states.
The simulated robots — a cheetah, ant and hopper — acquired transferable skills by being set loose to ‘explore’ a simulated 3D environment, trying out a series of randomized behaviors to learn a wide variety of skills. Unlike traditional reinforcement learning techniques, the simulated bots weren’t rewarded for particular behaviours, but instead some of these skills were later identified as useful for carrying out specific tasks, for example walking forward, and then fine-tuned to carry out that task.
The eventual application of the approach — dubbed Diversity is All You Need (DIAYN) — could include priming robots with skills that allow them to more rapidly learn how to carry out a wide range of tasks.
“We showed that DIAYN learns diverse skills for complex tasks, often solving benchmark tasks with one of the learned skills without actually receiving any task reward,” they note, adding there could be a variety of applications.
“From a perspective of creativity and education, the skills produced by DIAYN might be used by game designers to allow players to control complex robots and by artists to design dancing robots.”