We present PCA-Bench, a multimodal decision-making benchmark for evaluating
the integrated capabilities of Multimodal Large Language Models (MLLMs).
Departing from previous benchmarks focusing on simplistic tasks and individual
model capability, PCA-Bench introduces three complex scenarios: autonomous
driving, domestic robotics, and open-world games. Given task instructions and
diverse contexts, the model is required to seamlessly integrate multiple
capabilities of Perception, Cognition, and Action in a reasoning chain to make
accurate decisions. Moreover, PCA-Bench features error localization
capabilities, scrutinizing model inaccuracies in areas such as perception,
knowledge, or reasoning. This enhances the reliability of deploying MLLMs. To
balance accuracy and efficiency in evaluation, we propose PCA-Eval, an
automatic evaluation protocol, and assess 10 prevalent MLLMs. The results
reveal significant performance disparities between open-source models and
powerful proprietary models like GPT-4 Vision. To address this, we introduce
Embodied-Instruction-Evolution (EIE), an automatic framework for synthesizing
instruction tuning examples in multimodal embodied environments. EIE generates
7,510 training examples in PCA-Bench and enhances the performance of
open-source MLLMs, occasionally surpassing GPT-4 Vision (+3\% in decision
accuracy), thereby validating the effectiveness of EIE. Our findings suggest
that robust MLLMs like GPT4-Vision show promise for decision-making in embodied
agents, opening new avenues for MLLM research.

PCA-Bench 是一个用于评估多模态大型语言模型（MLLMs）综合能力的多模态决策基准，引入了三个复杂场景：自动驾驶、家庭机器人和开放世界游戏，并提出了误差定位能力和自动评估协议 PCA-Eval 对 10 种著名 MLLM 进行评估结果显示开源模型和 GPT-4 Vision 等强大专有模型之间存在显著性能差异，通过引入基于体验环境的自动框架 Embodied-Instruction-Evolution（EIE），在 PCA-Bench 中生成了 7,510 个训练示例，并提高了开源 MLLM 的性能，偶尔超越 GPT-4 Vision（+3％决策准确性），验证了 EIE 的有效性，发现 GPT4-Vision 之类的鲁棒 MLLM 对体验型代理的决策具有潜力，为 MLLM 研究开辟了新的道路。