Repeatability of Real World Training Experiments: A Case Study

  • Authors:
  • Dean F. Hougen;Paul E. Rybski;Maria Gini

  • Affiliations:
  • Department of Computer Science and Engineering, University of Minnesota, 200 Union St. S.E., Minneapolis, MN 55455-0159. hougen@cs.umn.edu;Department of Computer Science and Engineering, University of Minnesota, 200 Union St. S.E., Minneapolis, MN 55455-0159. rybski@cs.umn.edu;Department of Computer Science and Engineering, University of Minnesota, 200 Union St. S.E., Minneapolis, MN 55455-0159. gini@cs.umn.edu

  • Venue:
  • Autonomous Robots
  • Year:
  • 1999

Quantified Score

Hi-index 0.00

Visualization

Abstract

We present a case study of reinforcement learning on a real robot thatlearns how to back up a trailer and discuss the lessons learnedabout the importance of proper experimental procedure and design. Weidentify areas of particular concern to the experimental roboticscommunity at large. In particular, we address concerns pertinent torobotics simulation research, implementing learning algorithms on realrobotic hardware, and the difficulties involved with transferringresearch between the two.