Implementation of reinforcement learning environment for robotic arms supported by the CoppeliaSim simulator.
Tested on:
- Operation system - Ubuntu 20.04/22.04\
- CoppeliaSim 4.4.0
- Python 3.10
Used libraries:
- gym - framework for creating and managing Rl environments\
- stables-baselines3 - library containing latest reinforcement learning algorithms
- Pyrep - library for controlling the CoppeliaSim simulator
- download CoppeliaSim from the link above as a compressed file
- extract it on your computer
- add following lines to your bash profile:
export COPPELIASIM_ROOT=EDIT/ME/PATH/TO/COPPELIASIM/INSTALL/DIR
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$COPPELIASIM_ROOT
export QT_QPA_PLATFORM_PLUGIN_PATH=$COPPELIASIM_ROOT
- create a Python virtual environment via virtualenv command
- install all requirements from the file
requirements/dev.txt
For the last step, you may need to downgrade to setuptools==66. The issue is caused by using an old version of the gym package.
To see all possible options, run python -m roboticsrl.main --help
To install on a server, create Python virtual environment and install requirements from the file requirements/production.txt.
After this, simply run ./scripts/start-docker.sh.
All results of training and evaluation will be saved in $HOME/results/$CONTAINER_NAME directory.
Trained models with the best performance are saved in the models directory, and the positions directory
contains testing data for evaluating effectiveness.
To run demo of the trained models, simply start the roboticsrl/demo.py script.