A new project called Solaris is using Minecraft as a sandbox to train multi-agent AI systems in rich, open-ended environments. The model can generate synchronized first-person videos from two players in the same Minecraft world, covering actions such as fighting, building, mining, and exploring.
To build the dataset, the team created “Solaris Engine,” which runs two bots in a shared world and automatically records their gameplay. The result is a dataset of roughly 6.32 million frames per player, along with code that has been released publicly.
By simulating multi-player scenarios, Solaris could help researchers explore coordination, competition, and communication for both game AI and future robot swarms.
Comments
No comments yet. Be the first to share your thoughts!