Harvesting Common-sense Navigational Knowledge for Robotics from Uncurated Text Corpora

Nancy Fulda, Nathan Tibbetts, Zachary Brown, David Wingate
Proceedings of the 1st Annual Conference on Robot Learning, PMLR 78:525-534, 2017.

Abstract

As robotic systems are deployed into everyday situations, the need for abstract reasoning becomes more pronounced. The ideal robotic assistant should be able to understand verbal commands and work independently to fulfill human-prescribed goals, even if instructions are ambiguous or circumstances change. This paper presents a new algorithm for high-level reasoning based on Euclidean representations of words and their meanings. Rather than using ontologies or knowledge graphs, we model information about the world as a learned geometry of the contexts in which human beings tend to use each idea. Building on the analogy algorithms utilized by Mikolov et al., we perform mathematical operations on the vector space to infer responses to previously unseen problems, and apply our method to a sequence of semantic reasoning tasks in order to answer questions such as ‘Where can I find a dustpan?’, ‘Where do the crayons belong?’, and ‘What transportation method will bring me to the airport?’. Our Directional Scoring Method (DSM) returns a ranked list of possible responses, many of which are plausible answers to the query. Additionally, DSM’s top-ranked response is significantly more likely to be correct than the top-ranked responses of naive analogy estimations.

Cite this Paper

BibTeX
@InProceedings{pmlr-v78-fulda17a, title = {Harvesting Common-sense Navigational Knowledge for Robotics from Uncurated Text Corpora}, author = {Fulda, Nancy and Tibbetts, Nathan and Brown, Zachary and Wingate, David}, booktitle = {Proceedings of the 1st Annual Conference on Robot Learning}, pages = {525--534}, year = {2017}, editor = {Levine, Sergey and Vanhoucke, Vincent and Goldberg, Ken}, volume = {78}, series = {Proceedings of Machine Learning Research}, month = {13--15 Nov}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v78/fulda17a/fulda17a.pdf}, url = {https://proceedings.mlr.press/v78/fulda17a.html}, abstract = {As robotic systems are deployed into everyday situations, the need for abstract reasoning becomes more pronounced. The ideal robotic assistant should be able to understand verbal commands and work independently to fulfill human-prescribed goals, even if instructions are ambiguous or circumstances change. This paper presents a new algorithm for high-level reasoning based on Euclidean representations of words and their meanings. Rather than using ontologies or knowledge graphs, we model information about the world as a learned geometry of the contexts in which human beings tend to use each idea. Building on the analogy algorithms utilized by Mikolov et al., we perform mathematical operations on the vector space to infer responses to previously unseen problems, and apply our method to a sequence of semantic reasoning tasks in order to answer questions such as ‘Where can I find a dustpan?’, ‘Where do the crayons belong?’, and ‘What transportation method will bring me to the airport?’. Our Directional Scoring Method (DSM) returns a ranked list of possible responses, many of which are plausible answers to the query. Additionally, DSM’s top-ranked response is significantly more likely to be correct than the top-ranked responses of naive analogy estimations.} }
Endnote
%0 Conference Paper %T Harvesting Common-sense Navigational Knowledge for Robotics from Uncurated Text Corpora %A Nancy Fulda %A Nathan Tibbetts %A Zachary Brown %A David Wingate %B Proceedings of the 1st Annual Conference on Robot Learning %C Proceedings of Machine Learning Research %D 2017 %E Sergey Levine %E Vincent Vanhoucke %E Ken Goldberg %F pmlr-v78-fulda17a %I PMLR %P 525--534 %U https://proceedings.mlr.press/v78/fulda17a.html %V 78 %X As robotic systems are deployed into everyday situations, the need for abstract reasoning becomes more pronounced. The ideal robotic assistant should be able to understand verbal commands and work independently to fulfill human-prescribed goals, even if instructions are ambiguous or circumstances change. This paper presents a new algorithm for high-level reasoning based on Euclidean representations of words and their meanings. Rather than using ontologies or knowledge graphs, we model information about the world as a learned geometry of the contexts in which human beings tend to use each idea. Building on the analogy algorithms utilized by Mikolov et al., we perform mathematical operations on the vector space to infer responses to previously unseen problems, and apply our method to a sequence of semantic reasoning tasks in order to answer questions such as ‘Where can I find a dustpan?’, ‘Where do the crayons belong?’, and ‘What transportation method will bring me to the airport?’. Our Directional Scoring Method (DSM) returns a ranked list of possible responses, many of which are plausible answers to the query. Additionally, DSM’s top-ranked response is significantly more likely to be correct than the top-ranked responses of naive analogy estimations.
APA
Fulda, N., Tibbetts, N., Brown, Z. & Wingate, D.. (2017). Harvesting Common-sense Navigational Knowledge for Robotics from Uncurated Text Corpora. Proceedings of the 1st Annual Conference on Robot Learning, in Proceedings of Machine Learning Research 78:525-534 Available from https://proceedings.mlr.press/v78/fulda17a.html.

Related Material