Embedding and generation of indoor climbing routes with variational autoencoder

TL;DR

This paper presents a variational autoencoder approach to generate indoor climbing routes on MoonBoard, enabling automated route creation and potential assistance in route setting.

Contribution

It introduces a novel application of variational autoencoders for climbing route generation on MoonBoard, demonstrating high-quality route synthesis from latent space sampling.

Findings

Generated routes are of high quality and diverse.

22 routes uploaded for user review.

The method facilitates automated route setting.

Abstract

Recent increase in popularity of indoor climbing allows possible applications of deep learning algorthms to classify and generate climbing routes. In this work, we employ a variational autoencoder to climbing routes in a standardized training apparatus MoonBoard, a well-known training tool within the climbing community. By sampling the encoded latent space, it is observed that the algorithm can generate high quality climbing routes. 22 generated problems are uploaded to the Moonboard app for user review. This algorithm could serve as a first step to facilitate indoor climbing route setting.