Voyager minedojo

Voyager is an innovative agent that operates in the Minecraft environment and utilizes large language models. It is unique in that it can achieve lifelong learning without any human intervention. Voyager consists of three main components: an automatic curriculum, a skill library, and an iterative prompting mechanism. The automatic curriculum ensures that the agent explores its surroundings to the fullest by generating tasks based on the goal of discovering diverse things. This is similar to an in-context novelty search. The skill library stores and retrieves complex behaviors, making it easy for Voyager to access them when needed. By synthesizing simpler programs, Voyager can create more complex skills over time, thus enhancing its capabilities and avoiding catastrophic forgetting. The iterative prompting mechanism allows Voyager to learn from its mistakes and improve its skills by incorporating feedback from the environment, execution errors, and self-verification. Voyager interacts with GPT-4, a large language model, through blackbox queries, eliminating the need for fine-tuning of model parameters. In experiments, Voyager has demonstrated exceptional lifelong learning capabilities and proficiency in playing Minecraft. It has outperformed previous approaches by obtaining more unique items, covering longer distances, and achieving key milestones in the tech tree at a faster rate. Additionally, Voyager has shown superior generalization to novel tasks in new Minecraft worlds compared to other techniques. Through self-driven exploration, Voyager continually discovers new items and skills, surpassing baseline methods. Overall, Voyager combines the power of large language models with embodied learning to create an agent that can continuously explore, develop skills, and make novel discoveries in open-ended environments like Minecraft.