*********************************
There is now a CONTENT FREEZE for Mercury while we switch to a new platform. It began on Friday, March 10 at 6pm and will end on Wednesday, March 15 at noon. No new content can be created during this time, but all material in the system as of the beginning of the freeze will be migrated to the new platform, including users and groups. Functionally the new site is identical to the old one. webteam@gatech.edu
*********************************
Title: Game-Theoretic Planning for Parallel Play
Date: Friday, October 2nd, 2020
Time: 10:30 AM - 12:30 PM (EST)
Location: BlueJeans meeting (https://bluejeans.com/1191418749)
**Note: this proposal is remote-only**
Shray Bansal
Computer Science Ph.D. Student
School of Interactive Computing
Georgia Institute of Technology
Committee:
Dr. Charles L. Isbell (Advisor) – School of Interactive Computing, Georgia Institute of Technology
Dr. Ayanna Howard (Co-advisor) – School of Interactive Computing, Georgia Institute of Technology
Dr. Mark Riedl – School of Interactive Computing, Georgia Institute of Technology
Dr. Sonia Chernova– School of Interactive Computing, Georgia Institute of Technology
Abstract:
As more robots become integrated into human environments, they will find themselves sharing space with other independently acting agents like people or other robots. Sharing space often leads to conflict, even among humans, and managing this conflict well is key to task completion and successful integration. In this work, tasks, where humans and robots perform separate activities but interact due to shared space, are termed as parallel play. The goal of this research is to model such functions in the context of Human-Robot Interaction (HRI) and develop approaches that enable robots to perform their individual tasks efficiently while reducing the conflict with the people around them.
We have considered three parallel play tasks. First, in simulated driving, we model the interaction as a collaboration where the robot plans using a shared reward that combines both of their individual utilities. Balancing their goals here led to better task performance for both agents. Second, we develop interaction supporting actions and apply them to a close-proximity manipulation task between a robot and a human. These actions allow the robot to deliberately move the human’s objects in a way that reduces future goal conflict. Although they increase task completion time, they led to less interference with the human and increased the robot’s favorability as a coworker. These experiments show us that: (1) consideration of other’s goals is important for effective coordination and (2) human plans can be influenced to reduce the interference between the agents. We use these ideas to develop a game-theoretic technique that treats a parallel play task as a noncooperative game and computes the Nash equilibrium to plan agent actions. We apply it to a simulated close-proximity pick-and-place task and show that it can coordinate well with different kinds of agents with varying personalities.
Our plan for the proposed work aims to generalize this approach to an intersection navigation scenario. Here, we will model the car's high-level decision-making as parallel play to generate motion plans for the autonomous vehicle by finding the Nash equilibria. Our goal is to validate its applicability to this scenario by studying the interaction with human-driven cars. This will also allow us to test the feasibility of the approach to more than two agents.
