gym-package-template

This template repo was created following the Environment Generation guide from the Gymnasium Documentation. Created for and only tested on Ubuntu 20.04 & Windows 10 using Python 3.8.10.

A checklist for your RL project development with dynamic systems:

Dynamics Simulator

• Dynamics Class
• ODE Solver

Environment Generation

Base Environment

• Observation Space(s)
• Action Space(s)
• Data logging

SubEnvironment

Implementing subenvironments inheriting from the BaseEnv allows easy prototyping.

• __init__(self, render_mode, **kwargs): Specific environment configuration
• _get_obs(): Calculate the observation
• _get_info(): Log performance and info data
• _check_terminated(): Evaluate episode termination criteria
• _calculate_reward(): Implement the reward function
• reset(seed, options): Implement stochastic elements for environment reset after each episode
• step(action): Implement system dynamics, do environment changes
• render() + _render_frame(): Implement a renderer
• close(): Implement resource cleanup, close windows, threads, etc.

Deep Reinforcement Learning

• API compatible with gymnasium standard.
• Stable-Baselines3 integration.
• Multiprocessing with SubprocVecEnvs.
• Callback class for Tensorboard logging.

Installation

1. Clone this repository
2. Set up virtual environment using Python 3.8.10
3. Install the package locally

Add any additional required packages to setup.py.

Clone this repo

Show instructions

  git clone git@github.com:ThomasNLarsen/TTK29-RL-Lecture.git

Virtual environment

Show instructions

This repository was developed using Python 3.8.10.
Support for other versions is not guaranteed.

  python3 -m venv venv
  source venv/bin/activate

Install requirements

Show instructions

  python3 -m pip install -e .