PAI-Bench: A Comprehensive Benchmark For Physical AI
This provides a comprehensive benchmark for researchers in Physical AI to assess model capabilities, though it is incremental as it focuses on evaluation rather than new methods.
The paper tackles the problem of evaluating how well current models handle real-world dynamics in Physical AI by introducing PAI-Bench, a benchmark with 2,808 real-world cases, and finds that video generative models often lack physical coherence and multi-modal large language models perform poorly in forecasting and causal interpretation.
Physical AI aims to develop models that can perceive and predict real-world dynamics; yet, the extent to which current multi-modal large language models and video generative models support these abilities is insufficiently understood. We introduce Physical AI Bench (PAI-Bench), a unified and comprehensive benchmark that evaluates perception and prediction capabilities across video generation, conditional video generation, and video understanding, comprising 2,808 real-world cases with task-aligned metrics designed to capture physical plausibility and domain-specific reasoning. Our study provides a systematic assessment of recent models and shows that video generative models, despite strong visual fidelity, often struggle to maintain physically coherent dynamics, while multi-modal large language models exhibit limited performance in forecasting and causal interpretation. These observations suggest that current systems are still at an early stage in handling the perceptual and predictive demands of Physical AI. In summary, PAI-Bench establishes a realistic foundation for evaluating Physical AI and highlights key gaps that future systems must address.