Quality diversity search struggles to pick parents that produce lasting results
From Admission to Retention: Predicting Durable Offspring in Quality-Diversity Search
Neural and Evolutionary ComputingArtificial Intelligence
Summary
When trying to find many different ways a system can fail, it helps to keep the most useful failures around for long enough. The authors looked at ways to guess which failures will stick around longer based on clues before they reproduce. They found that some clues work differently when predicting entry versus long-term survival, and even when trying to favor parents that seem promising, the search does not get better and can get worse. This suggests it’s tricky to predict which behaviors will last in quality-diversity search.
What this means in practice
- •For roboticists: Improve diverse behavior exploration by understanding limitations of biasing parent selection in quality-diversity algorithms.
- •For software testers: Design automated testing suites that consider the durability of failure types rather than only their initial discovery.
Authors
Corrado Calamita, Matteo Mendicini, Luca Ferrari, Riccardo Bertamini
Abstract
Automated failure discovery can be framed as a diversity problem: the objective is to expose distinct ways a system fails rather than a single critical adversarial case. Yet uniform parent sampling provides no way to allocate search effort toward individuals likely to produce durable archive contributions. We study this credit assignment problem in an unstructured Quality-Diversity (QD) archive, testing whether signals available before an individual has reproduced distinguish offspring that merely enter the archive from those still retained generations later. Across a planar-arm benchmark and two robotic-locomotion tasks, the same signals are often associated with entry and persistence in opposite directions. In particular, lineage depth is the clearest such signal, and its persistence discrimination grows with population capacity, which is associated with lower turnover. Biasing parent selection toward the signals with the strongest measured association to survival nevertheless does not improve search: moderate bias matches uniform sampling, and strong bias reduces QD-score.