Cognitive Modeling for Human-Robot Value Soft Alignment
Keywords: Intention Recognition, Cognitive Modeling, Agent-Based Systems
Abstract: In the domain of human-robot symbiosis, it is of utmost importance for robots to display intelligent behavior. This encompasses the capability to deduce implicit information to predict the true intentions of humans. Traditionally, humans are perceived as flawless, with their decisions acting as the standards that robots should strive to align with. However, this raises a pertinent question: What if humans make mistakes? In this research, we present a unique task, termed ''value soft alignment''. This task involves the proactive identification of potentially hazardous or unsuitable human actions before the manifestation of their repercussions, and the provision of situationally appropriate suggestions. To facilitate this task, we have constructed a dataset for model training and testing, comprising two types of data: simulated human behaviors and collected human data. We propose a value-driven cognition model to represent the understanding of human behavior, followed by a two-stage method that consists of 1) the prediction of value-based long-term intention and 2) the comparison between the long-term intention and the short-term immediate action intention. Experimental results indicate that the value-driven cognition model can assist robots in comprehending human behavioral patterns over both long-term and short-term durations, and thus the robot can offer sensible recommendations for a majority of scenarios based on the consistency between long-term and short-term intentions of humans.
Supplementary Material: zip
Primary Area: general machine learning (i.e., none of the above)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 1725
Loading