Skip to content

Autoregressive policies for continuous control reinforcement learning

License

Notifications You must be signed in to change notification settings

kindredresearch/arp

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

39 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Autoregressive policies for continuous control reinforcement learning

This repository provides the implementation of autoregressive policies (ARPs) for continuous control deep reinforcement learning together with learning examples based on Open AI Baselines PPO and TRPO algorithms. The examples are provided for OpenAI Gym Mujoco environments and for Square sparse reward environment, discussed in the paper.

Tensorflow >= 1.12, OpenAI Baselines and OpenAI Gym are required to run learning examples. NumPy only is required to build and plot stationary AR processes.

Examples

  1. To generate and plot noise trajectories based on AR processes at different orders and smoothing parameter values

python ./examples/make_noise.py

  1. To run ARP with OpenAI Baselines PPO on a Square environment

python ./examples/run_square_ppo.py --dt 0.1 --p 3 --alpha 0.8 --num-timesteps=500000

  1. To run ARP with OpenAI Baselines PPO on a Mujoco environment

python ./examples/run_mujoco_ppo.py --env Reacher-v2 --p 3 --alpha 0.5 --num-timesteps=1000000

  1. To run ARP with OpenAI Baselines TRPO on a Mujoco environment

python ./examples/run_mujoco_trpo.py --env Reacher-v2 --p 3 --alpha 0.5 --num-timesteps=1000000

Reference

Autoregressive Policies for Continuous Control Deep Reinforcement Learning.
Dmytro Korenkevych, A. Rupam Mahmood, Gautham Vasan, James Bergstra. arXiv preprint, 2019.
paper | video

About

Autoregressive policies for continuous control reinforcement learning

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages