tower-mind-agent-benchmark

【效率与协作】Evaluate LLM agent capabilities using tower defense game environment with multimodal observations (pixel, text, structured state). Benchmark reveals critical agent limitations: inadequate planning validation, inflexible decision-making, and inefficient action use. Demonstrates significant performance gap between current LLMs and human experts, provid…

独立 GitHub 快照 · 自动分类 · 以原作者文档为准