With the increasing integration of service robots into assistive technologies, there is a need to reason about the boundaries and scope of these robots' autonomy, such as when they should merely react to their environment, when they should make proactive decisions, and when they should override commands. In most existing research, the definition of a ``good'' assistive robot is one that is compliant with respect to the commands it is given. Two recent papers challenge this perspective, and describe scenarios where a system might choose to rebel against a command or disobey its handler due to a deep understanding of the handler's intentions. This paper provides a comparative discussion about these two papers and how they together create a more comprehensive framework for assistive robots that can override commands.

PDF

In ASIMOV workshop as part of the International Conference on Intelligent Robots and Systems (IROS), Virtual, November 2021.

@inproceedings{ASIMOV2021-REUTH, title={Intelligent Disobedience and AI Rebel Agents in Assistive Robotics}, author={Reuth Mirsky and Peter Stone}, booktitle={ASIMOV workshop as part of the International Conference on Intelligent Robots and Systems (IROS)}, month={November}, address={Virtual}, url="http://nn.cs.utexas.edu/?ASIMOV2021-REUTH", year={2021} }