We present Large Model Navigation (LM-Nav) --- a method that combines the strengths of large, pre-trained models of language, images, and visual navigation, for the task of embodied instruction following.

"LM-Nav: Robotic Navigation with Large Pre-Trained Models of Language, Vision, and Action"
Dhruv Shah*, Blazej Osinski*, Brian Ichter, Sergey Levine
Berkeley AI Research (UC Berkeley), University of Warsaw, Robotics at Google

Presented at Conference on Robot Learning (CoRL) 2022, Auckland NZ
Project Page: [ Ссылка ]
arXiv pre-print: [ Ссылка ]

Timeline:
-------------
00:00 Introduction
00:17 Problem Setup
00:38 Method Overview
02:53 LM-Nav in the Real-World
04:12 Another Real-World Experiment
04:39 Disambiguating Textual Instructions
05:56 The End

свернуть