Artificial intelligence has been advancing rapidly, with the latest algorithms producing very impressive outputs. However, even the highest performing algorithms struggle to guarantee that high-level constraints are satisfied. This becomes a particular challenge when considering environments where constraints can change over time, for example when a user adds or removes constraints based on previously generated content. To address this gap we are introducing the first QDA-VC competition at AIIDE 2026 - a competition that centers on the design of controllable algorithms that adapt to changing constraint landscapes.
The Quality-Diversity Algorithm with Variable Constraints (QDA-VC) benchmark provides a suite of problem spaces, procedural personas (simulated users), and baseline algorithms to test how well quality-diversity algorithms (a subtype of genetic algorithms) can adapt to constraints given a variety of situations. The problem spaces include two procedural content generation (PCG) spaces: generating logic grid puzzles and Lode Runner levels; and one optimization space: the traveling thief problem. The procedural personas are able to add (and remove) constraints within these bases, for example forcing a Lode Runner level to have a minimum number of gold. Competitors will be asked to create algorithms that perform well across all procedural personas and problem spaces - including a secret problem space that will only be revealed after the competition has ended. The baseline algorithms can be used as a starting point for developing new algorithms, as well as a way to measure progress.
The QDA-VC is a great opportunity, especially for students, to not only contribute to an emerging research space but to develop their own expertise in artificial intelligence. This competition is unique not only in its focus on controllability but in its application to procedural content generation across multiple problem spaces. Given how novel this space is, participants have the opportunity to have their efforts make big impacts.
Test and enhance your personal algorithm development skills
Contribute to the research community and efforts towards controllable interactive genetic algorithms
During the last day of the AIIDE-26 conference, there will be a presentation of the results of the competition during which all submissions (especially the winner) will be recognized.
Submissions open August 1st
Submissions due September 8th
Results announced November 13th
The full competition details, as well as the code-base, are available on GitHub: https://github.com/fiabot/VariableConstraintGA/tree/competition
Please reach out to the organizer, Fiona Shyne, with any questions or concerns at: shyne.f@northeastern.edu.