Programming robots effectively remains a challenge for small businesses due to the high ongoing costs of robot programming experts. What is missing is a user-friendly software system such as a natural language-enabled cognitive assistant for developing robot programs that (1) does not require any particular training before it can be used, and (2) allows for natural instruction dialogues that let human operators develop programs interactively. In this paper, we introduce such as system, specifically the cognitive robotic TRACS architecture which enables industrial robots to learn from human teachers through natural language dialogues novel tasks that can be executed immediately. We briefly describe the core elements of the architecture and present a sorting task example to showcase how a task can be first instructed and later modified during task execution, all in multiple different spoken natural languages.
@inproceedings{scheutzetal24arci,
title={Dialogue-Based Task Instructions and Modifications for Industrial Robots},
author={M. Scheutz and B. Oosterveld and J. Peterson and E. Wyss},
year={2024},
booktitle={Proceedings of ARCI'24},
url={https://hrilab.tufts.edu/publications/scheutzetal24arci.pdf}
}