Demo 2020: Backchannel Generation with Batch RL

Training Socially Engaging Robots: Modeling Backchannel Behaviors with Batch Reinforcement Learning, N. Hussain, E. Erzin, T. M. Sezgin, Y. Yemez

Sample video recording with the RB policy

Sample video recording with the RL policy

This demo contains a sample video snippet from one of the recordings from IEMOCAP dataset. In the original recording, two participants are interacting with each other. We have prepared the video by replacing the participant on the right with an animated version of Furhat. The smiles/nods of the robot in the animation are triggered by an offline trained RL policy. The three RL policies are the baselines DQN-agent and NFQ-agent, and the proposed SRDQN-agent.