Designing interaction for human-machine collaboration in multi-agent scheduling

Abstract

In the field of multi-agent task scheduling, there are many algorithms that are capable of minimizing objective functions when the user is able to specify them. However, there is a need for systems and algorithms that are able to include user preferences or domain knowledge into the final solution. This will increase the usability of algorithms that would otherwise not include some characteristics desired by the end user but are highly optimal mathematically. We hypothesize that allowing subjects to iterate over solutions while adding allocation and temporal constraints would allow them to take advantage of the computational power to solve the temporal problem while including their preferences. No statistically significant results were found that supported that such algorithm is preferred over manually solving the problem among the participants. However, there are trends that support the hypothesis. We found statistically significant evidence (p=0.0027), that subjects reported higher workload when working with Manual Mode and Modification Mode rather than Iteration Mode and Feedback Iteration Mode. We propose changes to the system that can provide guidance for future design of interaction for scheduling problems.