SAGA: Scene-Aware, Goal-Evolving Agents for Long-Horizon CivRealm Strategy Planning

2026-06-29

AI Summary

This work addresses key challenges in complex strategy games—namely, situational blindness, contextual overload, and shallow cross-game learning—by proposing a multi-agent strategic decision-making framework. The approach integrates semantic scene graphs to model spatial relationships, employs a tool-augmented, domain-specific planner for on-demand reasoning, and introduces a dual-timescale causal replay mechanism to drive goal evolution. Evaluated on the FreeCiv benchmark, the method significantly outperforms baseline approaches, achieving the highest average civilization score with the lowest variance, demonstrating superior infrastructure-building capability, reducing reasoning overhead by 27%, and exhibiting consistent performance gains across five consecutive matches. These results validate the framework’s efficiency and evolvability in dynamic strategic environments.

Abstract

Long-horizon strategic planning in complex strategy games demands concurrent reasoning across multiple decision domains under imperfect information and sparse reward. Existing LLM-based agents suffer from three systematic failures: scene blindness from raw tile coordinates, context overflow and domain coupling from monolithic state dumps, and shallow cross-game learning that treats each episode in isolation. We present SAGA, an LLM multi-agent framework with three mechanisms each directly targeting one class of failure: (i) a Map-Semantic Scene Graph that encodes typed spatial relations among game entities into per-unit natural-language context, resolving spatial blindness without global token inflation; (ii) a Tool-Augmented Planner that pulls fine-grained domain state on demand and dispatches per-domain directives to dedicated specialist controllers, eliminating context overflow, domain coupling, and mechanical constraint violations; and (iii) a Dual-Horizon Feedback Loop that combines periodic within-game goal generation with structured cross-game causal post-mortem, enabling principled strategic evolution without manual reward engineering. Evaluated on FreeCiv, SAGA attains the highest mean civilization score -- the environment's sole sparse objective reward -- with lower variance than the two strongest baselines, and is the only method that significantly surpasses every baseline on infrastructure construction, the resource axis most readily sacrificed under multi-objective conflict. It outscores the two strongest baselines in most head-to-head games while cutting output tokens (the dominant decoding cost) by 27%. Equipped with the cross-game evolution module, SAGA reaches the highest end-of-chain score across five successive episodes. Ablation studies confirm that each architectural component contributes independently to this advantage.

Problem

• long-horizon planning
• scene awareness
• cross-game learning
• sparse reward
• domain coupling

Innovation

• Map-Semantic Scene Graph
• Tool-Augmented Planner
• Dual-Horizon Feedback Loop
• LLM Multi-Agent Framework
• Long-Horizon Strategic Planning