Learning Object Properties Using Robot Proprioception via Differentiable Robot-Object Interaction

2024-10-04

arXiv.org

AI Summary

This work addresses the problem of inferring physical properties of manipulated objects—such as mass, inertia tensor, and elastic modulus—using only robot proprioceptive data (joint encoder measurements), without object-mounted sensors, vision systems, or external measurement tools. We propose an end-to-end differentiable robot–object interaction modeling framework that integrates differentiable physics simulation with gradient-based system identification to jointly estimate these parameters. To our knowledge, this is the first approach enabling full physical property calibration without any sensing on the object side, while supporting plug-and-play deployment on arbitrary articulated robots. Evaluated on low-cost hardware, the method achieves high-precision real-time estimation within seconds on a laptop: mass estimation error remains below 3.2%, and elastic modulus error below 5.7%. The approach significantly advances perception capabilities for sensorless, lightweight robotic systems.

Technology Category

• Intelligent Robots: Manipulation
• Computer Vision: Low Level & Physics-based Vision
• Machine Learning: Calibration & Uncertainty Quantification

Application Category

• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Differentiable simulation has become a powerful tool for system identification. While prior work has focused on identifying robot properties using robot-specific data or object properties using object-specific data, our approach calibrates object properties by using information from the robot, without relying on data from the object itself. Specifically, we utilize robot joint encoder information, which is commonly available in standard robotic systems. Our key observation is that by analyzing the robot's reactions to manipulated objects, we can infer properties of those objects, such as inertia and softness. Leveraging this insight, we develop differentiable simulations of robot-object interactions to inversely identify the properties of the manipulated objects. Our approach relies solely on proprioception -- the robot's internal sensing capabilities -- and does not require external measurement tools or vision-based tracking systems. This general method is applicable to any articulated robot and requires only joint position information. We demonstrate the effectiveness of our method on a low-cost robotic platform, achieving accurate mass and elastic modulus estimations of manipulated objects with just a few seconds of computation on a laptop.

Problem

• Calibrates object properties using robot proprioception.
• Infers object properties like inertia and softness via robot reactions.
• Uses differentiable simulations for robot-object interaction analysis.

Innovation

• Uses robot proprioception for object property calibration
• Differentiable simulation identifies object properties indirectly
• Requires only joint position data, no external sensors