This project investigated how humans perceive collaborative behaviour in artificial agents, in the context of guided sound synthesis.
Our work focused on Reinforcement Learning agents (RL), whose goal is to take actions in an environment so as to maximize some notion of reward. We were interested in creating a collaborative behaviour in agents in an interactive context where humans would communicate rewards following agents’ actions.
We thus led a controlled experiment where we asked participants to guide three types of agents using only feedback, in two sound synthesis environment. Our results shift standard RL assumptions of optimal behaviour to interactive RL requirements that critically integrate the agent’s exploration behaviour in addition to its reaching a goal.
The project was developed with Frédéric Bevilacqua and Baptiste Caramiaux in collaboration with the ISMM group of IRCAM and the ex)situ group of LRI (INRIA), in the context of the Sorbonne Université Doctorate in Computing.
Paper at SMC (2018)