Evolving Complex Environments in Evolution Gym using Large Language Models

Creating systems capable of generating virtually infinite variations of complex and novel behaviour without predetermined goals or limits is a major challenge in the field of AI. This challenge has been addressed through the development of several open-ended algorithms that can continuously generate new and diverse behaviours, such as the Enhanced-POET algorithm for co-evolving environments and agent behaviour. One of the challenges with existing methods however, is that they struggle to generate complex environments. In this work, we propose a method in which Evolution Gym environments are created using Large Language Models (LLMs). By fine-tuning a LLM with text representations of Evolution Gym environments, and captions that describe the environment, we were able to generate complex and diverse environments using natural language. We then implement this environment generation method into the Enhanced-POET algorithm, replacing the CPPNs that are typically used. We found that not only could the LLM produce a diverse range of environments, but compared to CPPNs, the LLMs allowed for a 34% increase in the performance gain of EnhancedPOET. This increased performance suggests that the agents were able to learn a more diverse set of skills by training on more complex environments.