Increasing GPS Localization Accuracy Using Reinforcement Learning

The authors propose a reinforcement learning-based framework to increase the GPS localization accuracy. The framework does not make any assumptions on the GPS device hardware parameters or motion models, nor does it require infrastructure-based reference locations. The proposed reinforcement learning model learns an optimal strategy to make ``corrections'' on raw GPS observations. The model uses an efficient confidence-based reward mechanism, which is independent of the geolocation, thereby enabling the model to be implemented anywhere. The authors use a map matching-based regularization term to reduce the variance of the reward return. The authors implement the asynchronous advantage actor-critic (A3C) algorithm, a parallel training protocol, for training their model. A3C facilitates short training sessions and provides more robust performance. The authors' real-world experiments demonstrate that using A3C the authors can obtain an optimal policy in about 800 steps in a training session with 4 threads. The authors provide visualization results to demonstrate the performance of the proposed framework by comparing the original GPS observations with model outputs.

  • Supplemental Notes:
    • This paper was sponsored by TRB committee ABJ70 Standing Committee on Artificial Intelligence and Advanced Computing Applications.
  • Corporate Authors:

    Transportation Research Board

    ,    
  • Authors:
    • Zhang, Ethan
    • Masoud, Neda
  • Conference:
  • Date: 2019

Language

  • English

Media Info

  • Media Type: Digital/other
  • Features: Figures; References;
  • Pagination: 6p

Subject/Index Terms

Filing Info

  • Accession Number: 01698032
  • Record Type: Publication
  • Report/Paper Numbers: 19-05157
  • Files: TRIS, TRB, ATRI
  • Created Date: Dec 7 2018 9:44AM