Discipline: Computer Sciences and Information Management
Subcategory: Computer Science & Information Systems
Christopher Banks - Norfolk State University
Co-Author(s): Joseph Kim and Julie Shah, Massachusetts Institute of Technology, Cambridge, MA
Prior work in human in-the-loop planning has focused on the type of interaction where the human acts as a supervisor. The user typically accepts/rejects generated solutions, tweaks parameters, or redefines the whole problem specification. We are interested in a type of interaction where the user participates in the actual plan generation process. Through an iterative planning framework, we seek to reduce the plan computation time and increase the user’s understanding of the plan. A state-of-the-art automated planner was tested on benchmark domains in order to create a baseline for automated plan generation. Then, we conducted a human subject study to analyze user decision-making processes. The findings of this research support that user-provided sub-goals can increase plan efficiency for complex planning tasks. This project enabled a better understanding of the types of suggestions given by the user and their effectiveness on varying levels of planning complexity.
Funder Acknowledgement(s): This study was funded by the MIT Summer Research Program (MSRP).
Faculty Advisor: Julie Shah, julie_a_shah@csail.mit.edu
Role: I was responsible for creating the experiment used to determine users ability to create plans. I then tested the total time to create the plan and plan cost across two different types of domains. I compared these results to that of an automated planner that completed the same tasks. I also plotted the percent difference in plan cost and total time for humans to complete the tasks in comparison with the automated planner. This was accomplished in MATLAB with box and whisker plots to highlight the median value and outliers.