Evolving Balancing Controllers for Biped Characters in Games

Christopher Schinkel Carlsen, George Palamas

Research output: Contribution to book/anthology/report/conference proceedingArticle in proceedingResearchpeer-review

Abstract

This paper compares two approaches to physics based, balancing systems, for 3D biped characters that can react to dynamic environments. The first approach, based on the concept of proprioception, use a neuro-controller to define the position and orientation of the joints involved in the motion. The second approach
use a self-adaptive Proportional Derivative (PD) controller along with a neural network. Both neural networks were trained using a Genetic Algorithm (GA). The study showed that both approaches were capable of achieving balance and the GA proved to work well as a search strategy for both the neuro-controller and the PD-controller. The results also showed that the neuro-controller performed better but the PD-controller was more flexible and capable to recover under external disturbances such as wind drag and momentary collisions with objects.
Original languageEnglish
Title of host publicationEvolving Balancing Controllers for Biped Characters in Games
PublisherSpringer
Publication dateJun 2019
Publication statusPublished - Jun 2019

Fingerprint

Controllers
Derivatives
Genetic algorithms
Neural networks
Drag
Physics

Keywords

  • Neural network applications
  • Virtual agents
  • Procedural Animation
  • Animation

Cite this

Carlsen, C. S., & Palamas, G. (2019). Evolving Balancing Controllers for Biped Characters in Games. In Evolving Balancing Controllers for Biped Characters in Games Springer.
Carlsen, Christopher Schinkel ; Palamas, George. / Evolving Balancing Controllers for Biped Characters in Games. Evolving Balancing Controllers for Biped Characters in Games. Springer, 2019.
@inproceedings{70d751f03a884efdbe6401894c595372,
title = "Evolving Balancing Controllers for Biped Characters in Games",
abstract = "This paper compares two approaches to physics based, balancing systems, for 3D biped characters that can react to dynamic environments. The first approach, based on the concept of proprioception, use a neuro-controller to define the position and orientation of the joints involved in the motion. The second approachuse a self-adaptive Proportional Derivative (PD) controller along with a neural network. Both neural networks were trained using a Genetic Algorithm (GA). The study showed that both approaches were capable of achieving balance and the GA proved to work well as a search strategy for both the neuro-controller and the PD-controller. The results also showed that the neuro-controller performed better but the PD-controller was more flexible and capable to recover under external disturbances such as wind drag and momentary collisions with objects.",
keywords = "Neural network applications, Virtual agents, Procedural Animation, Animation",
author = "Carlsen, {Christopher Schinkel} and George Palamas",
year = "2019",
month = "6",
language = "English",
booktitle = "Evolving Balancing Controllers for Biped Characters in Games",
publisher = "Springer",
address = "Germany",

}

Carlsen, CS & Palamas, G 2019, Evolving Balancing Controllers for Biped Characters in Games. in Evolving Balancing Controllers for Biped Characters in Games. Springer.

Evolving Balancing Controllers for Biped Characters in Games. / Carlsen, Christopher Schinkel; Palamas, George.

Evolving Balancing Controllers for Biped Characters in Games. Springer, 2019.

Research output: Contribution to book/anthology/report/conference proceedingArticle in proceedingResearchpeer-review

TY - GEN

T1 - Evolving Balancing Controllers for Biped Characters in Games

AU - Carlsen, Christopher Schinkel

AU - Palamas, George

PY - 2019/6

Y1 - 2019/6

N2 - This paper compares two approaches to physics based, balancing systems, for 3D biped characters that can react to dynamic environments. The first approach, based on the concept of proprioception, use a neuro-controller to define the position and orientation of the joints involved in the motion. The second approachuse a self-adaptive Proportional Derivative (PD) controller along with a neural network. Both neural networks were trained using a Genetic Algorithm (GA). The study showed that both approaches were capable of achieving balance and the GA proved to work well as a search strategy for both the neuro-controller and the PD-controller. The results also showed that the neuro-controller performed better but the PD-controller was more flexible and capable to recover under external disturbances such as wind drag and momentary collisions with objects.

AB - This paper compares two approaches to physics based, balancing systems, for 3D biped characters that can react to dynamic environments. The first approach, based on the concept of proprioception, use a neuro-controller to define the position and orientation of the joints involved in the motion. The second approachuse a self-adaptive Proportional Derivative (PD) controller along with a neural network. Both neural networks were trained using a Genetic Algorithm (GA). The study showed that both approaches were capable of achieving balance and the GA proved to work well as a search strategy for both the neuro-controller and the PD-controller. The results also showed that the neuro-controller performed better but the PD-controller was more flexible and capable to recover under external disturbances such as wind drag and momentary collisions with objects.

KW - Neural network applications

KW - Virtual agents

KW - Procedural Animation

KW - Animation

M3 - Article in proceeding

BT - Evolving Balancing Controllers for Biped Characters in Games

PB - Springer

ER -

Carlsen CS, Palamas G. Evolving Balancing Controllers for Biped Characters in Games. In Evolving Balancing Controllers for Biped Characters in Games. Springer. 2019