Toward Cooperative Team-diagnosis in Multi-robot Systems

  • Authors:

  • Michael D.M. Kutzer;Mehran Armand;David H. Scheid;Ellie Lin;Gregory S. Chirikjian

  • Affiliations:

  • Johns Hopkins, University Applied Physics, Laboratory 11100, Johns Hopkins Road, Laurel, MD 20723;Johns Hopkins, University Applied Physics, Laboratory 11100, Johns Hopkins Road, Laurel, MD 20723;Johns Hopkins, University Applied Physics, Laboratory 11100, Johns Hopkins Road, Laurel, MD 20723;Carnegie-Mellon University, Robotics Institute 5000 Forbes Avenue, Pittsburgh, Pennsylvania;Johns Hopkins University, Department of Mechanical Engineering 3400 North Charles Street, Baltimore, Maryland 21218

  • Venue:
  • International Journal of Robotics Research
  • Year:
  • 2008

Quantified Score

Hi-index 0.00

Visualization

Abstract

Research in man-made systems capable of self-diagnosis and self-repair is becoming increasingly relevant in a range of scenarios in which in situ repair/diagnosis by a human operator is infeasible within an appropriate time frame. In this paper, we present an approach to the multi-robot team diagnosis problem that utilizes gradient-based training of multivariate Gaussian distributions. We then evaluate this approach using a testbed involving modular mobile robots, each assembled from four electromechanically separable modules. The diagnosis algorithm is trained on data obtained from two sources: (1) a computer model of the system dynamics and (2) experimental runs of the physical prototypes. Tests were then performed in which a fault was introduced in one robot in the testbed and the diagnostic algorithm was queried. The results show that the state predicted by the diagnostic algorithm performed well in identifying the fault state in the case when the model was trained using the experimental data. Limited convergence was also demonstrated using training data from an imperfect dynamic model and low data sampling frequencies.