Train reinforcement learning agents on the grand challenge of civilization building.

Terra Nova is an RL environment based on Civilization V, a 4X strategy game. Terra Nova is designed to be a testbest for general intelligence research, requiring agents to reason over a large set of diverse information streams and control hundreds of heterogeneous endpoints simultaneously. Terra Nova is a comprehensive challenge environment, where many canonical RL challenges (e.g., credit assignment, partial observability, enormous action space, etc) occur simultaneously.

Project by Trevor McInroe