Publication

On the application of CMA-ES to biped walk

Technical Report (2017)

IRI code

IRI-TR-17-03

File

Download the digital copy of the doc pdf document

Authors

Abstract

This project studies the applicability of Covariance Matrix Adaptation Evolution Strategies to enable a biped robot to walk with no previous knowledge. While most work on the subject is based on Genetic Algorithms, we believe that CMA-ES shows indeed great potential. Our work focuses on the Darwin robot, although it can be easily extrapolated to other humanoids. The evaluation of the solutions' goodness is performed via simulation to keep the real robot away from potential harm.

Categories

artificial life, humanoid robots, learning (artificial intelligence).

Author keywords

biped walking, evolutionary algorithms, genetic algorithms, evolution strategies

Scientific reference

A. Suárez, C. Miralles and S. Medina. On the application of CMA-ES to biped walk. Technical Report IRI-TR-17-03, Institut de Robòtica i Informàtica Industrial, CSIC-UPC, 2017.