Hacker News new | comments | show | ask | jobs | submitlogin
Learning State Abstractions for Long-Horizon Planning (bair.berkeley.edu)
43 points by headalgorithm 13 days ago | hide | past | web | 1 comment | favorite

Techniques like these seem very exciting for me in terms of making AI more general-purpose.

This type of thing indicates to me that we will see robots that can effectively learn and perform a variety of tasks in the real world in the relatively near future. Within years rather than decades. I mean it's not a fully functional system and doesn't solve all problems, but it shows that there is a type of "functional-checking" approach to abstraction that could work, hopefully in multiple contexts.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | DMCA | Apply to YC | Contact