Challenges in Policy Learning for British Sign Language Users in an Example of Navigation-Based Task

In the light of natural human interactions, this paper identi€es the lack of multimodality in current policy learning approaches for human-robot/human-computer interactions. Larger part of policies, trained for the navigation-based tasks are trained on using either textual or natural language inputs, which are unimodal. Œis paper suggests and formalises British Sign Language as an input to the o„ine example-based policy learning algorithm. Œe policy for how to follow directional instructions from a human to a machine is learned through reinforcement. Currently, the training environment is being set-up and the datasets that will be used for the o„ine policy learning are being analysed. Œe novelty of this project is the use of sign languages in policy learning for navigationbased task that exceeds existing research in computing that uses sign languages as well as advances the research on multimodal human-robot/human-computer interactions. ACM Reference format: . 2017. Challenges in Policy Learning for British Sign Language Users in an Example of Navigation-Based Task. In Proceedings of 19th ACM International Conference on Multimodal Interaction, Glasgow, Scotland UK, November 2017