View in EDS

Bootstrapped Neuro-Simulation as a method of concurrent neuro-evolution and damage recovery.

Saved in:
Bibliographic Details
Title: Bootstrapped Neuro-Simulation as a method of concurrent neuro-evolution and damage recovery.
Authors: Leonard, Brydon A.1 (AUTHOR) s214036685@mandela.ac.za, du Plessis, Mathys C.1 (AUTHOR) MC.duPlessis@mandela.ac.za, Woodford, Grant W.1 (AUTHOR) s205014224@mandela.ac.za
Source: Robotics & Autonomous Systems. Feb2020, Vol. 124, pN.PAG-N.PAG. 1p.
Subjects: Closed loop systems, Evolutionary computation, Dog training
Abstract: Bootstrapped Neuro-Simulation (BNS) is a method of concurrent simulator and robot controller evolution. The algorithm requires little domain knowledge and no pre-investigation data gathering. Additionally, it bridges the reality gap effectively, rapidly evolves functional controllers, and recovers from damage automatically. In this paper, the first evidence of the ability of BNS to evolve closed-loop controllers is shown; in this case to solve a light-following problem. The algorithm is then evaluated for its damage recovery ability for these closed-loop controllers and shown to be very effective, with only minor adaptations. • Bootstrapped Neuro-Simulation is shown to evolve closed-loop controllers. • The algorithm starts in a random state and evolves controllers in minutes. • The algorithm can recover from damage to the robot. • Sliding windows of training data offer the best performance for damage recovery. [ABSTRACT FROM AUTHOR]
Copyright of Robotics & Autonomous Systems is the property of Elsevier B.V. and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database: Engineering Source
Description
Abstract:Bootstrapped Neuro-Simulation (BNS) is a method of concurrent simulator and robot controller evolution. The algorithm requires little domain knowledge and no pre-investigation data gathering. Additionally, it bridges the reality gap effectively, rapidly evolves functional controllers, and recovers from damage automatically. In this paper, the first evidence of the ability of BNS to evolve closed-loop controllers is shown; in this case to solve a light-following problem. The algorithm is then evaluated for its damage recovery ability for these closed-loop controllers and shown to be very effective, with only minor adaptations. • Bootstrapped Neuro-Simulation is shown to evolve closed-loop controllers. • The algorithm starts in a random state and evolves controllers in minutes. • The algorithm can recover from damage to the robot. • Sliding windows of training data offer the best performance for damage recovery. [ABSTRACT FROM AUTHOR]
ISSN:09218890
DOI:10.1016/j.robot.2019.103398