Spatial Reasoning in LLM Game Agents: Impact of Causal Context and Multi-Step Planning

2026-07-22

AI Summary

This work addresses the limited performance of large language models in complex game tasks, primarily attributed to insufficient spatial reasoning capabilities. To investigate this, the authors propose an enhanced approach integrating causal prompting with multi-step planning and introduce a three-tiered spatial navigation benchmark based on the General Video Game AI (GVGAI) framework for systematic evaluation across model scales and reasoning paradigms. Experiments conducted on the Qwen3 model family incorporate chain-of-thought reasoning and variable planning horizons. Results demonstrate that enabling chain-of-thought significantly improves localization accuracy, causal context effectively increases win rates, and multi-step planning enhances overall performance while reducing per-step response latency. The study reveals the synergistic role of causal reasoning and planning mechanisms in optimizing agents’ spatial reasoning abilities.

Abstract

LLM-based game agents often perform poorly on more complex tasks. This work examines whether these failures are linked to limited spatial reasoning and evaluates whether causal prompt augmentation and multi-step planning can improve win-rates while managing response latency. Using the open-source Qwen3 model family, we conduct experiments across varying model scales, reasoning modes, and planning horizons. We further introduce a focused GVGAI benchmark consisting of three custom games with five difficulty levels to isolate spatial navigation. The evaluation follows two paradigms: an initial ``positioning experiment'' to test an agent's ability to find its exact coordinates, and a study of game-play success. Our results show that while larger models with an enabled thinking mode identify their positions more accurately, overall performance in coordinate matching remains limited for smaller models. Win rates decrease as game levels and layout complexity increase, validating the benchmark's difficulty scaling. Integrating causal context into the prompts tends to improve the agents' success rates, particularly for bigger models. While enabling thinking mode and longer planning horizons significantly improve performance, multi-step planning further reduces mean per-step response times, offering a practical trade-off between reasoning depth and execution speed.

Problem

• spatial reasoning
• LLM game agents
• complex tasks
• coordinate matching
• game-play success

Innovation

• spatial reasoning
• causal prompt augmentation
• multi-step planning
• LLM game agents
• GVGAI benchmark