/dec-hdrqn

Primary LanguagePython

Multi-Agent Deep RL for Independent Learners

Supported Features

  • DRQN (with LSTM)
  • Distributed Experience Trajectories
  • Quantile Networks (both implicit and explicit)
  • Double Learning
  • Hysteretic (supports annealing, toggled using hynamic_h param)
  • Time Difference Likelihood!
  • Risk Distortion

Example

python main.py -n_quant 8 --env cmotp1 --likely 1 --distort_type wang

will train agents in cmotp1 environment with TDL turned on and wang as distortion function and using 8 quantile samples. See more parameters in main.py.

Included environments

  • Meeting-In-A-Grid - capture_target
  • CMOTP - cmotp1, cmotp2, cmotp3
  • Climb Game - climb_game
  • Catch-Pig - pig