researchers teach robots to learn just like humans - by trial and error

a team of researchers from china’s zhenjiang university and the university of edinburgh has created a new method to train robots inspired in the way humans learn — through trial and error. this new approach distances itself, for example, from boston dynamic’s robots which are fed with precise lines of code that dictate every movement. on a recent paper published in the journal science robotics, the team presents yueying, a robot dog that learns how to get up after being knocked down.

‘the AI approach is very different in the sense that it captures experience, which the robot has tried hundreds of thousands of times, or even millions of times,’ university of edinburgh roboticist zhibin li said to wired. ‘so in the simulated environment, I can create all possible scenarios. I can create different environments or different configurations. for example, the robot can start in a different pose, such as lying down on the ground, standing, falling over, and so on.’

to train the robot, the team created eight algorithms for each skill — walking, balancing, steering, etc. each algorithm was rewarded with virtual points once they successfully completed a task and deployed from them if they failed, leading them to master the skill. an additional network was created to control the eight algorithms, acting as a kind of brain or coach. ‘the coach or the captain will tell who is doing what, or who should do work together, at which time, so all experts can collaborate together as a whole team, and this drastically improves the capability of skills,’ li comments.

if you can’t get enough of this, watch boston dynamics’ robots celebrate the end of 2020 with a perfectly choreographed dance party.

the eight algorithms controlled by the coach

project info:

paper name: multi-expert learning of adaptive legged locomotion

researchers: chuanyu yang, kai yuan, qiuguo zhu, wanming yu and zhibin li

journal: science robotics

via: wired