Terra Nova: A Comprehensive Challenge Environment for Intelligent Agents

TL;DR

Terra Nova is a new comprehensive challenge environment for reinforcement learning that combines multiple complex challenges in a single setting, aiming to evaluate integrated, long-term reasoning abilities of agents.

Contribution

The paper introduces Terra Nova, a unified environment designed to test reinforcement learning agents on multiple interacting challenges simultaneously, unlike existing multitask benchmarks.

Findings

Terra Nova presents a complex environment with multiple RL challenges.

It emphasizes integrated reasoning over isolated task performance.

The environment aims to push the development of more capable RL agents.

Abstract

We introduce Terra Nova, a new comprehensive challenge environment (CCE) for reinforcement learning (RL) research inspired by Civilization V. A CCE is a single environment in which multiple canonical RL challenges (e.g., partial observability, credit assignment, representation learning, enormous action spaces, etc.) arise simultaneously. Mastery therefore demands integrated, long-horizon understanding across many interacting variables. We emphasize that this definition excludes challenges that only aggregate unrelated tasks in independent, parallel streams (e.g., learning to play all Atari games at once). These aggregated multitask benchmarks primarily asses whether an agent can catalog and switch among unrelated policies rather than test an agent's ability to perform deep reasoning across many interacting challenges.