Header logo is am

Towards Full System Linear Quadratic Regulators for Humanoid Control


Conference Paper



Robots that are to locomote in a human like fashion requirecontrol of high degree of freedom (DOF) motions potentiallycoupled in a complex way. It remains challenging to expressthe task objective in an intuitive way and simultaneously generatefeedback gains guaranteeing some level of optimality.In response to this, a number of different simplified modelshave been developed to highlight different aspects of the humanoidâ??sdynamics that are important for specific tasks. Ashort list of some of the models used to represent a humanoidinclude the cart-table, double inverted pendulum, reactionmass pendulum, and automatically generated task specific reducedmodels [4]. These simplified models make planningeasier but come at the cost of modelling error and limitationson motion. In addition, one is tasked with finding mappingsbetween the full system to the reduced system. These mappingscan potentially destroy the intuition surrounding the useof the simplified model as they may not always behave as expected.By working with the full dynamics, one obtains anincrease in generality, accuracy, and eliminates the need formappings.

Author(s): Mason, S. and Righetti, L. and Schaal, S.
Book Title: Proceedings of Dynamic Walking
Year: 2014

Department(s): Autonomous Motion, Movement Generation and Control
Bibtex Type: Conference Paper (inproceedings)

Address: Zurich, Switzerland
Cross Ref: p10591
Note: clmc
URL: http://www-clmc.usc.edu/publications/M/mason-DW2014.pdf

Links: PDF


  title = {Towards Full System Linear Quadratic Regulators for Humanoid Control},
  author = {Mason, S. and Righetti, L. and Schaal, S.},
  booktitle = {Proceedings of Dynamic Walking},
  address = {Zurich, Switzerland},
  year = {2014},
  note = {clmc},
  crossref = {p10591},
  url = {http://www-clmc.usc.edu/publications/M/mason-DW2014.pdf}