This is a configurable Gynmasium environment that implements one or more agents moving in a plane. The software was originally inspired by a game for kids called Sapientino, but it has been extended in various ways.
Each agent moves on a 2D environment, where each cell can be coloured or blank. When a robot is on a coloured cell, it can execute a beep action, meaning it has visited the cell (this is meant to represent any interaction with the current location).
The environment is compliant with the Gymnasium APIs.
Agents There can be one or more agents in the same map. Each agent has its own action space.
Actions There are three default action spaces. The first allows the agent to move in the four cardinal directions. The second requires the agent to rotate by 90°, then move in discrete steps. The third instead allows the agent to accellerate and decelerate both in the angular and linear coordinates. This last modality does not implement a grid-world environment. For these actions, and how to implement your own, you can see gym_sapientino/core/actions.py.
Observations The Sapientino class has a dictionary observation space that contains all the current information. For personalizing the observation space you can subclass the Features class in gym_sapientino/wrappers/observations.py. We provide discrete and continuous features wrappers.
Rewards It is possible to specify per-step rewards, but for general reward functions, the user should wrap this environment.
Map The map can be easily configured using ASCII strings. For example, combining
|P bB g |
| bp G r|
|G pg |
| rpG PB|
|rP Bg b|
with ContinuousCommand generates:
For a more complete documentation of how to use the environment, and the available options see this notebook.
Install with pip:
pip install gym_sapientino
Or, for a more updated version, install from github:
git clone https://github.com/whitemech/gym-sapientino.git
cd gym-sapientino
pip install .
- Clone the repo:
git clone https://github.com/whitemech/gym-sapientino.git
cd gym-sapientino- Install Poetry
- Optionally select the Python version:
poetry env use python3.9- Install with development dependencies:
poetry installTo run tests: tox
Please look at the tox.ini file for the full list of supported commands and tests.
gym-sapientino is released under the GNU General Public License v3.0 or later (GPLv3+).
Copyright 2019-2020 Marco Favorito, Roberto Cipollone, Luca Iocchi
The code is largely inspired by RLGames
If you want to use this environment in your research, please consider citing this conference paper:
@inproceedings{Giacomo2019FoundationsFR,
title={Foundations for Restraining Bolts: Reinforcement Learning with LTLf/LDLf Restraining Specifications},
author={Giuseppe De Giacomo and L. Iocchi and Marco Favorito and F. Patrizi},
booktitle={ICAPS},
year={2019}
}