Evolved Controllers for Simulated Locomotion

  • Authors:
  • Brian F. Allen;Petros Faloutsos

  • Affiliations:
  • University of California, Los Angeles, USA 90024;University of California, Los Angeles, USA 90024

  • Venue:
  • MIG '09 Proceedings of the 2nd International Workshop on Motion in Games
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

We present a system for automatically evolving neural networks as physics-based locomotion controllers for humanoid characters. Our approach provides two key features: (a) the topology of the neural network controller gradually grows in size to allow increasingly complex behavior, and (b) the evolutionary process requires only the physical properties of the character model and a simple fitness function. No a priori knowledge of the appropriate cycles or patterns of motion is needed.