MBRL

Model Based Reinforcement Learning

This module has only one function call:

MBRLAgent(Models, Environments, AgentParameter, ModelParameter)

This function combines model free agents with model learning.

This module depends mainly on:

• RLTypes
• DDPG
• DQN
• NODEDynamics
• ODERNNDynamics
• Conda
• PyCall
• Gymnasium, this is a slightly adapted version. The original package can be found here

Installation

In the Julia REPL, run

using Pkg
Pkg.add("Conda")
Pkg.add("PyCall")
Pkg.add(url="https://github.com/SvenDuve/MBRL.jl")
Pkg.add(url="https://github.com/SvenDuve/RLTypes.jl")

Then again within julia,

using Conda
using PyCall
Conda.pip_interop(true)
Conda.add("wheel")
Conda.add("box2d-py")
Conda.add("pygame")
Conda.pip("install", "git+https://github.com/SvenDuve/Gymnasium")

Bring package into scope with

using MBRL

Example usage

Once the above is installed, the following code can be used to train an agent:

using MBRL
using RLTypes
MBRLAgent(NODEModel(), Pendulum(), AgentParameter(training_episodes=500), ModelParameter())

Caution, this code will take a long time to run.