Update #3

The past few weeks, we’ve been focused on building an AI for the game. We’re using ML-agents to try to build an agent that will provide a compelling competitive experience. We’ve experimented with various reinforcement learning best practices, e.g., tuning hyperparameters, adjusting rewards architectures, and experimenting with training data features.

As the graphic shows, the AI looks to be unplayable. Hopefully we can share better progress soon.

Leave a Reply