keiohta / torchrl

Reinforcement Learning and Inverse Reinforcement Learning using PyTorch

torchrl

torchrl is a deep reinforcement and inverse reinforcement learning tool using PyTorch. This project is a reimplementation of tf2rl, a deep reinforcement learning algorithms using Tensorflow 2.x, using pytorch. Hence, all the kudos go to tf2rl.

Algorithms

Overview

Algorithm	Category	Check Results
BC	Imitation Learning	x
DDPG	Model-free Off-policy RL	✓
DQN	Model-free Off-policy RL	Mid-way
GAIL	Imitation Learning	x
SAC	Model-free Off-policy RL	✓

Algorithm Reference

• Model-free Off-policy RL
  • DDPG: Continuous control with deep reinforcement learning
  • DQN: Deep Reinforcement Learning with Double Q-learning
  • SAC: Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor
• Imitation Learning
  • BC: Model-Free Imitation Learning with Policy Optimization
  • GAIL: Generative Adversarial Imitation Learning

Installation

Currently only available for local installation

# support only through pip

$ git clone [e-mail]:sff1019/torchrl.git
$ cd torchrl
$ pip install -r requirements.txt
$ pip install -e .

Reference

• tf2rl: The base of this project

Quick commands (removing soon)

# to render on server
$ xvfb-run -s "-screen 0 1400x900x24" python [file_name].py