Aligning Text and Embodied Environments for Interactive Learning    

Mohit Shridhar Xingdi (Eric) Yuan Marc-Alexandre Côté
Yonatan Bisk   Adam Trischler Matthew Hausknecht

ALFWorld contains interactive TextWorld environments (Côté et. al) that parallel embodied worlds in the ALFRED dataset (Shridhar et. al). The aligned environments allow agents to reason and learn high-level policies in an abstract space before solving embodied tasks through low-level actuation.   

Paper on ArXiv» GitHub» ALFRED» TextWorld»

You are in the middle of a room. Looking quickly around you, you see a safe 1, a shelf 4, a drawer 2, a bed 1, a drawer 1, a shelf 5, a shelf 2, a sidetable 2, a shelf 3, a drawer 3, a shelf 1, a sidetable 1, a desk 1, and a garbagecan 1. 

Your task is to: examine an alarmclock with the desklamp. 

> go to desk 1
You arrive at loc 8. On the desk 1, you see a pen 1, a bowl 1, a alarmclock 2, a pencil 2, a pencil 3, a creditcard 3, a book 1, a alarmclock 3, a keychain 3, and a book 2.

> take alarmclock 2 from desk 1
You pick up the alarmclock 2 from the desk 1.

> go to sidetable 2
You arrive at loc 1. On the sidetable 2, you see a desklamp 1, and an alarmclock 1.

> use desklamp 1
You won!