REPLICATING IMPULSE-BASED PHYSICS ENGINE USING CLASSIC NEURAL NETWORKS

RareÈ™-Cristian Ifrim (), Patricia Penariu () and Costin-Anton Boiangiu ()
Additional contact information
RareÈ™-Cristian Ifrim: Politehnica University of Bucharest, Romania
Patricia Penariu: Politehnica University of Bucharest, Romania
Costin-Anton Boiangiu: Politehnica University of Bucharest, Romania

Journal of Information Systems & Operations Management, 2021, vol. 15, issue 2, 175-186

Abstract: The high costs for creating and using traditional simulators induced both by technical effort, which extends over a long period, but also by the need for permanent updating during this period, to improve accuracy by using a limited range of settings, bring to the fore an alternative, namely: data-based methods for physical simulation; they are a much more attractive option for interactive applications in terms of their ability to trade precomputation and memory footprint for better performance while running. Also trained physics engines might come to offer better simulations as the networks can be configured to take as input data from real-world measurements, thus it can combine the best from the real-time simulation engines that emphases on fast simulations but do not offer a real-world-like simulation, and high accuracy simulation engines and targets real-world simulations but require high computational resources. This paper aims to construct a neural network that learns how the impulses between two objects react when they make a contact, by using an already implemented physics engine for generating the training datasets and to compare the results of the trained engine versus the original one. Although this has been done successfully, the proposed neural network managing to score a prediction rate with values between 55 and 89% depending on the test “scenario†, improvements can be made to increase performance and to obtain a suitable accuracy (over 90%, even 95%), thus achieving the goal of completely replacing the physics engine.

Date: 2021
References: View complete reference list from CitEc
Citations:

Downloads: (external link)
http://www.rebe.rau.ro/RePEc/rau/jisomg/WI21/JISOM-WI21-A15.pdf (application/pdf)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:rau:jisomg:v:15:y:2021:i:2:p:175-186

Access Statistics for this article

More articles in Journal of Information Systems & Operations Management from Romanian-American University Contact information at EDIRC.
Bibliographic data for series maintained by Alex Tabusca ().