PhysQuantAgent: An Inference Pipeline of Mass Estimation for Vision-Language Models

TL;DR

This paper introduces PhysQuantAgent, a framework leveraging vision-language models and a new benchmark dataset to improve real-world object mass estimation for robotic manipulation, demonstrating significant accuracy gains through visual prompting methods.

Contribution

The work presents a novel framework and benchmark for physical property estimation using VLMs, incorporating visual prompting techniques to enhance mass inference accuracy in real-world scenarios.

Findings

Visual prompting improves mass estimation accuracy

The benchmark dataset enables realistic evaluation of physical reasoning

Spatial reasoning enhances VLM capabilities for physical inference

Abstract

Vision-Language Models (VLMs) are increasingly applied to robotic perception and manipulation, yet their ability to infer physical properties required for manipulation remains limited. In particular, estimating the mass of real-world objects is essential for determining appropriate grasp force and ensuring safe interaction. However, current VLMs lack reliable mass reasoning capabilities, and most existing benchmarks do not explicitly evaluate physical quantity estimation under realistic sensing conditions. In this work, we propose PhysQuantAgent, a framework for real-world object mass estimation using VLMs, together with VisPhysQuant, a new benchmark dataset for evaluation. VisPhysQuant consists of RGB-D videos of real objects captured from multiple viewpoints, annotated with precise mass measurements. To improve estimation accuracy, we introduce three visual prompting methods that enhance the input image with object detection, scale estimation, and cross-sectional image generation to help the model comprehend the size and internal structure of the target object. Experiments show that visual prompting significantly improves mass estimation accuracy on real-world data, suggesting the efficacy of integrating spatial reasoning with VLM knowledge for physical inference.