Candidate: Daiwei Lin
Title: Interactive Behaviour Generation for Human-Robot Interaction Based on Reinforcement Learning
Date: December 18, 2019
Time: 10:00am
Place: EIT 3142
Supervisor(s): Kulic, Dana - Gorbet, Robert B. (Knowledge Integration)
Abstract:
This thesis is part of the research activities of the Living Architecture System Group (LASG). LASG develops art sculptures combining concepts of architecture, art, and electronics which allow occupants to interact with immersively. The primary goal of this research is to investigate e_ective human-robot interaction behaviours using reinforcement learning. Despite the complexity in current LASG systems, the question of how engaging the human designed behaviours of interactive sculptures can be still remains unanswered.
In this thesis, reinforcement learning is used to exploit potentials of human designed behaviours in engaging occupants. The algorithm is examined in a simulation environment built in Unity. The LAS system was simulated and two simplified human visitor models are designed for the tests. Three adaptive behaviour modes and two exploration methods are compared. We show that reinforcement learning algorithms can learn to increase engagement by adapting to visitors preferences and parameter noise is more suitable for such application usage.
A field study was conducted based on LASGs installation, Transforming Space, in Royal Ontario Museum (ROM) from June 2nd to October 8th, 2018. The experiment was conducted in a natural setting where no constraints are imposed on visitors and group interaction is accommodated. Experimental results demonstrate that learning on top of human designed pre-scripted behaviours
(PLA) is better at keeping visitors engagement than only using pre-scripted behaviours (PB). Visitor survey data shows a higher rating of PLA than PB in Likeablity and interactivity.