/Variational-Recurrent-Models

Codes for the study "Variational Recurrent Models for Solving Partially Observable Control Tasks", published as a conference paper at ICLR 2020 (https://openreview.net/forum?id=r1lL4a4tDB)

Primary LanguagePythonMIT LicenseMIT

Variational Recurrent Models for Solving Partially Observable Control Tasks

Codes for the study "Variational Recurrent Models for Solving Partially Observable Control Tasks", published as a conference paper at ICLR 2020 (https://openreview.net/forum?id=r1lL4a4tDB)

Language:

  • Python3.6

Library dependencies:

  • PyTorch (1.1.0, CPU version)
  • numpy (1.16.4)
  • scipy (1.3.0)
  • gym (0.12.5)
  • roboschool (for the robotic control tasks)
  • docopt (0.6.2)

To run an experiment:

python run_experiment.py run --env=PendulumV --steps=50000 --seed=0 --render

The program will run and save the results as .mat files at './data/' when finished (you can read it using scipy.io.loadmat in Python). Also, the trained agent will be saved as a PyTorch Module.

List of available env_names (V: velocities only, P: no velocities):

  • Pendulum
  • PendulumV
  • PendulumP
  • CartPole
  • CartPoleV
  • CartPoleP
  • Hopper
  • HopperV
  • HopperP
  • Ant
  • AntV
  • AntP
  • Walker2d
  • Walker2dP
  • Walker2dV
  • Sequential

Cognitive Neurorobotics Research Unit, Okinawa Institute of Science and Technology Graduate University

https://groups.oist.jp/cnru