Novel Model May Help Ro­bots Nav­i­gate More Like Hu­mans

The Day After - - COFFEE HOUSE -

As ro­bots may strug­gle with nav­i­ga­tional con­cepts, a team of re­searchers from the Mas­sachusetts In­sti­tute of Tech­nol­ogy (MIT) have de­vised a new way to help them nav­i­gate en­vi­ron­ments more as hu­mans do. The novel mo­tion-plan­ning model lets ro­bots de­ter­mine how to reach a goal by ex­plor­ing the en­vi­ron­ment, ob­serv­ing other agents, and ex­ploit­ing what they’ve learned be­fore in sim­i­lar sit­u­a­tions.

“Just like when play­ing chess, these de­ci­sions branch out un­til (the ro­bots) find a good way to nav­i­gate. But un­like chess play­ers, (the ro­bots) ex­plore what the fu­ture looks like with­out learn­ing much about their en­vi­ron­ment and other agents,” said co-au­thor An­drei Barbu from the var­sity. “The thou­sandth time they go through the same crowd is as com­pli­cated as the first time,” Barbu added.

Newspapers in English

Newspapers from India

© PressReader. All rights reserved.