Date of Award
evolutionary algorithms, genetic algorithms, bouldering, rock climbing, map-elites, neural networks, grading, moonboard
The challenge of utilizing artificial intelligence to generate indoor rock climbing routes with a specific grade is an interesting and unsolved problem due to its complexity and subjectivity. We use MAP-Elites, an evolutionary, quality-diversity algorithm, in conjunction with GradeNet  to produce a set of disjoint MoonBoard climbing routes that sufficiently challenge a climber without exceeding their physical and technical limitations. We evaluate these routes through visual a assessment survey by climbers as well as an in-person study in which climbers attempt to climb the generated routes. While our algorithm generally performs well in producing complete or near-complete archives of diverse climbs at every difficulty level as assessed by GradeNet, they fall short when it comes to in person trials. Additionally, the data from user surveys, while supporting the claims of Duh and Chang  about GradeNet's superiority to human grad- ing ability, is inconclusive in determining the success of our algorithm. These results leave open the path for future work to leverage the relative success of quality-diversity while accounting for the shortcomings of route quality and difficulty present in our system's design.
Tyebkhan, Daniel, "Evolving Difficulty Targeted Bouldering Routes" (2023). Honors Theses. 2752.