« Co-Explorers » are software agents that uses deep reinforcement learning to support interactive design space exploration.

They allow people to explore parametric design spaces by only communicating arbitrary preferences to the machine. A deep reinforcement learning algorithm is used to autonomously explore the space while interactively learning the user’s tastes—in a « co-exploration » workflow.

The current version of the software implements different user controls over agent’s exploration, along with a history that maps the space with user preferences. It is coded in Python; the OSC protocol supports connection between the deep reinforcement learning model (coded with the TensorFlow library) and parametric environments. Future developments may be released as a Python library and a Max/MSP object. It has been evaluated and used in workshops and practiced within the ægo artwork.

computer programming
interaction design
data design


The project was developed with Bavo Van Kerrebroeck and Frédéric Bevilacqua in collaboration with the ISMM group of IRCAM, in the context of the Sorbonne Université Doctorate in Computing.

Available on GitHub