eduzhai > Applied Sciences > Engineering >

Augmenting Differentiable Simulators with Neural Networks to Close the Sim2Real Gap

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 4 pages

Abstract: We present a differentiable simulation architecture for articulatedrigid-body dynamics that enables the augmentation of analytical models withneural networks at any point of the computation. Through gradient-basedoptimization, identification of the simulation parameters and network weightsis performed efficiently in preliminary experiments on a real-world dataset andin sim2sim transfer applications, while poor local optima are overcome througha random search approach.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...