Skip to main content

Evaluating RL Tricks using Sweeps

Created on November 2|Last edited on November 10
Watched Antonin Raffin pres about RL and real robots:


-> implemented tricks from the pres:
  • continuity cost
  • sGDE
  • HistoryWrapper (to avoid breaking markov assumption)
also fixed a few issues:
  • action limiter seemed to be causing issues -> removed it (might want to keep parts of it?)
  • limits for speed seemed to be too low. try different values