InteRACT: Transformer Models for Human Intent Prediction Conditioned on Robot Actions

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Kedia, Kushal, Bhardwaj, Atiksh, Dan, Prithwish, Choudhury, Sanjiban
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916270205566976
author Kedia, Kushal
Bhardwaj, Atiksh
Dan, Prithwish
Choudhury, Sanjiban
author_facet Kedia, Kushal
Bhardwaj, Atiksh
Dan, Prithwish
Choudhury, Sanjiban
contents In collaborative human-robot manipulation, a robot must predict human intents and adapt its actions accordingly to smoothly execute tasks. However, the human's intent in turn depends on actions the robot takes, creating a chicken-or-egg problem. Prior methods ignore such inter-dependency and instead train marginal intent prediction models independent of robot actions. This is because training conditional models is hard given a lack of paired human-robot interaction datasets. Can we instead leverage large-scale human-human interaction data that is more easily accessible? Our key insight is to exploit a correspondence between human and robot actions that enables transfer learning from human-human to human-robot data. We propose a novel architecture, InteRACT, that pre-trains a conditional intent prediction model on large human-human datasets and fine-tunes on a small human-robot dataset. We evaluate on a set of real-world collaborative human-robot manipulation tasks and show that our conditional model improves over various marginal baselines. We also introduce new techniques to tele-operate a 7-DoF robot arm and collect a diverse range of human-robot collaborative manipulation data, which we open-source.
format Preprint
id arxiv_https___arxiv_org_abs_2311_12943
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle InteRACT: Transformer Models for Human Intent Prediction Conditioned on Robot Actions
Kedia, Kushal
Bhardwaj, Atiksh
Dan, Prithwish
Choudhury, Sanjiban
Robotics
Artificial Intelligence
Machine Learning
Multiagent Systems
In collaborative human-robot manipulation, a robot must predict human intents and adapt its actions accordingly to smoothly execute tasks. However, the human's intent in turn depends on actions the robot takes, creating a chicken-or-egg problem. Prior methods ignore such inter-dependency and instead train marginal intent prediction models independent of robot actions. This is because training conditional models is hard given a lack of paired human-robot interaction datasets. Can we instead leverage large-scale human-human interaction data that is more easily accessible? Our key insight is to exploit a correspondence between human and robot actions that enables transfer learning from human-human to human-robot data. We propose a novel architecture, InteRACT, that pre-trains a conditional intent prediction model on large human-human datasets and fine-tunes on a small human-robot dataset. We evaluate on a set of real-world collaborative human-robot manipulation tasks and show that our conditional model improves over various marginal baselines. We also introduce new techniques to tele-operate a 7-DoF robot arm and collect a diverse range of human-robot collaborative manipulation data, which we open-source.
title InteRACT: Transformer Models for Human Intent Prediction Conditioned on Robot Actions
topic Robotics
Artificial Intelligence
Machine Learning
Multiagent Systems
url https://arxiv.org/abs/2311.12943