PBench

PBench is a benchmark that evaluates the physical realism and action alignment of video-based world models for robotic manipulation, utilizing a curated dataset of three million manipulation clips with physics-aware annotation.

Papers using PBench (5)

ABot-PhysWorld: Interactive World Foundation Model for Robotic Manipulation with Physics Alignment2026Qwen-RobotWorld Technical Report: Unifying Embodied World Modeling through Language-Conditioned Video Generation2026Perish or Flourish? A Holistic Evaluation of Large Language Models for Code Generation in Functional Programming2026Refining Critical Thinking in LLM Code Generation: A Faulty Premise-based Evaluation Framework2025PyBench: Evaluating LLM Agent on various real-world coding tasks2024