ALFWorld aligns text-based and embodied visual environments so agents can learn abstract policies in TextWorld that transfer to better performance on ALFRED tasks than visual-only training.
Specifically, embeddings are initialized by pre-trained 768-dimensional BERT embeddings (Sanh et al., 2019)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2020 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
ALFWorld: Aligning Text and Embodied Environments for Interactive Learning
ALFWorld aligns text-based and embodied visual environments so agents can learn abstract policies in TextWorld that transfer to better performance on ALFRED tasks than visual-only training.