pyRDDLGym-rl

Author: Mike Gimelfarb

This repository provides wrappers for deep reinforcement learning algorithms (i.e. Stable Baselines 3 and RLlib) to work with pyRDDLGym.

[!NOTE]
If your environment has differentiable dynamics and a differentiable reward, try the gradient-based JAX planner.

Contents

• Installation
• Running the Basic Examples
  • Stable Baselines 3
  • RLlib
• Creating an Environment

Installation

To run the basic examples you will need pyRDDLGym>=2.0, rddlrepository>=2.0 and one of the supported reinforcement learning frameworks:

• stable-baselines3>=2.2.1
• ray[rllib]>=2.9.2

You can install this package, together with all of its requirements via pip:

pip install stable-baselines3  # need one of these two
pip install -U "ray[rllib]"
pip install rddlrepository pyRDDLGym-rl

Running the Basic Examples

Stable Baselines 3

To run the stable-baselines3 example, navigate to the install directory of pyRDDLGym-rl, and type:

python -m pyRDDLGym_rl.examples.run_stable_baselines <domain> <instance> <method> <steps> <learning_rate>

where:

• <domain> is the name of the domain in rddlrepository, or a path pointing to a domain.rddl file
• <instance> is the name of the instance in rddlrepository, or a path pointing to an instance.rddl file
• <method> is the RL algorithm to use [a2c, ddpg, dqn, ppo, sac, td3]
• <steps> is the (optional) number of samples to generate from the environment for training, and
• <learning_rate> is the (optional) learning rate to specify for the algorithm.

RLLib

To run the RLlib example, from the install directory of pyRDDLGym-rl, type:

python -m pyRDDLGym_rl.examples.run_rllib <domain> <instance> <method> <iters>

where:

• <domain> is the name of the domain in rddlrepository, or a path pointing to a domain.rddl file
• <instance> is the name of the instance in rddlrepository, or a path pointing to an instance.rddl file
• <method> is the RL algorithm to use [dqn, ppo, sac]
• <iters> is the (optional) number of iterations of training

Creating an Environment

You can create an environment wrapper to use with your own RL implementations, or a package that is not currently supported by us:

import pyRDDLGym
from pyRDDLGym_rl.core.env import SimplifiedActionRDDLEnv
env = pyRDDLGym.make("domain name", "instance name", base_class=SimplifiedActionRDDLEnv)

This creates an instance of gymnasium.Env in which the action space is simplified by concatenating all continuous, discrete and boolean action fluents into single tensors. You can then use this environment as you would in pyRDDLGym.