Recent progress in reinforcement learning (RL) has started producing
generally capable agents that can solve a distribution of complex environments.
These agents are typically tested on fixed, human-authored environments. On the
other hand, quality diversity (QD) optimization has been