Inference of Mechanical Properties of Dynamic Objects through Active Perception

Wagner, Nikolaus and Cielniak, Grzegorz (2021) Inference of Mechanical Properties of Dynamic Objects through Active Perception. In: Towards Autonomous Robotic Systems Conference (TAROS), September 8-10, 2021.

Full content URL: https://doi.org/10.1007/978-3-030-89177-0_45

Documents
Inference of Mechanical Properties of Dynamic Objects through Active Perception
Accepted Manusript
[img] PDF
p2.13-TAROS2021paper24.pdf - Whole Document
Restricted to Repository staff only until 31 October 2022.

286kB
Item Type:Conference or Workshop contribution (Paper)
Item Status:Live Archive

Abstract

Current robotic systems often lack a deeper understanding of their surroundings, even if they are equipped with visual sensors like RGB-D cameras. Knowledge of the mechanical properties of the objects in their immediate surroundings, however, could bring huge benefits to applications such as path planning, obstacle avoidance & removal or estimating object compliance.
In this paper, we present a novel approach to inferring mechanical properties of dynamic objects with the help of active perception and frequency analysis of objects' stimulus responses. We perform FFT on a buffer of image flow maps to identify the spectral signature of objects and from that their eigenfrequency. Combining this with 3D depth information allows us to infer an object's mass without having to weigh it.
We perform experiments on a demonstrator with variable mass and stiffness to test our approach and provide an analysis on the influence of individual properties on the result. By simply applying a controlled amount of force to a system, we were able to infer mechanical properties of systems with an eigenfrequency of around 4.5 Hz in about 2 s. This lab-based feasibility study opens new exciting robotic applications targeting realistic, non-rigid objects such as plants, crops or fabric.

Keywords:active perception, image flow, frequency analysis
Subjects:H Engineering > H671 Robotics
G Mathematical and Computer Sciences > G740 Computer Vision
Divisions:College of Science > School of Computer Science
Related URLs:
ID Code:46646
Deposited On:29 Sep 2021 09:43

Repository Staff Only: item control page