Route-setters create routes for indoor climbing combining technical craft with an artistic representation of real rock climbing moves. Indoor routes need to be changed very often and, with indoor climbing recent boom, the demand for route-setters is rapidly increasing.
Can we use Deep Generative Models to automate route-setting?
In a first exploratory stage we will limit ourselves to routes on a fixed-topology input, such as a MoonBoard. Plotted against a grid of lettered and numbered coordinates, each Moonboard hold is rotated and set in a specific location. Climbers can share problems with fellow MoonBoard users around the globe via a dedicated app. With thousands of MoonBoards available in climbing gyms around the world, a large quantity of labelled routes (~37k) is available online. We propose to automate climbing routes generation by using Deep Conditional Generative Models .
 Sohn et al. "Learning Structured Output Representation using Deep Conditional Generative Models." NIPS 2015
Back to the project list.
The candidate should have programming experience, ideally in Python. Previous experience with Deep Learning is a plus. No climbing experience needed, but it would be fun to test the generated routes out!
20% Theory, 40% Implementation, 40% Research and Experiments