Candidate: Lingheng Meng
Title: Engaging Behavior Generation from Human Preferences In A Large Scale Interactive System: A Simulation Experiment
Date: April 10, 2023
Time: 9:00 AM
Place: REMOTE ATTENDANCE
Supervisor(s): Kulic, Dana (Adjunct) - Gorbet, Rob
Abstract:
Physical agents that can autonomously generate engaging, life-like behavior will lead to more responsive and interesting robots and other autonomous systems. In order to generate engaging behaviors, the autonomous system must first be able to estimate its human partners’ engagement level, then take actions to maximize the estimated engagement. In this seminar, we take a Living Architecture System (LAS), an architecture scale interactive system capable of group interaction through distributed embedded sensors and actuators, as a testbed, and apply Preference Learning (PL) and Deep Reinforcement Learning (DRL) by inducing a reward function from human preference then treating the induced reward signal as the estimate of engagement in order to automatically generate engaging behavior. We will start with introducing the overall framework and the high-level learning algorithm, followed by describing three types of preference teachers, i.e., simulated preference, constrained human preference and unconstrained expert preference, investigated in this work. At the experiment part, we will give a brief introduction of the simulation environment, and show the results on the three types of preference teachers with a special focus on the unconstrained expert preference