AI Daily - Artificial Intelligence News

View Original

Four-Legged Robot That Taught Itself How To Walk

A robot now has the walking capabilities of Bambi, which is more impressive than it sounds.  

Researchers at Google, the Georgia Institute of Technology, and UC Berkeley published a paper detailing the success of building a robot which used AI to teach itself how to walk. They nicknamed the little four-legged robot “Rainbow Dash.”  

Four-legged robot nicknamed “Rainbow Dash,” able to walk without assistance from humans. Image Credit: Google

They used an artificial intelligence technique called deep reinforcement learning, which combines aspects from two different kinds of AI, deep learning and reinforcement learning. Reinforcement learning is the use of an algorithm which learns how to perform a task through trial-and-error, where it receives rewards and punishments depending on how close it is to achieving the task that’s set. Deep learning allows systems to preview and evaluate raw input data from their environments. Deep reinforcement learning was programmed on the task of learning to walk.  

Head of locomotion at Google and researcher Jie Tan, stated that the research took approximately a year to complete. 

The robot started rocking backwards and forwards before using trial and error to understand that it could propel itself forward by bending its legs in the correct sequence.  

The AI system is capable of teaching itself to reliably walk across all three different terrains: flat ground, a soft mattress and a doormat with crevices. Researchers started first testing Rainbow Dash on a flat surface, where it figured out how to walk within 1.5 hours. It was then tested on slightly more challenging surfaces, a memory-foam mattress followed by a doormat with lots of crevices. The robot took 5.5 hours to figure out how to walk backwards and forwards on the mattress with less time on the doormat, which only took 4.5 hours. 

Once it had learned to walk, researchers were able to plug in a game console controller which took control of the robot and manoeuvred it forward, backward, left and right using the movements it learnt. 

The main benefit of a robot which does not require a manual input is that the AI framework can learn how to walk on a variety of surfaces without the need to program each necessary gait individually. 

The researchers hope to now test the learning technique on robots in a range of other situations and improve the systems so it could be used outside the lab in real-world conditions. 

Rainbow Dash was not entirely bereft of human supervision, researchers still had to intervene when the robot accidentally left the space it was mean to be learning in, and when it fell over – although an algorithm was used to help the robot to stand back up whenever it fell over and to stop it from wandering out of its space.