To train reinforcement learning agents for robotics applications requires access to accurate simulation environment. V-REP is popular robotics simulation environment with wide range of pre-built artifacts available, however, the Python API for V-REP is rusty. This project contains two reference environments which implement OpenAI gym like interface using V-REP.
Line follower using infrared (IR) sensors.
Obstacle avoidance using proximity sensors.
Create a Python 3 virtual environment and install the dependencies.
pip install -r requirements.txt
First, install V-REP 3.6. Once V-REP is installed, you would need to link V-REP binaries and Python API files.
On MacOS and Linux simply run the following script,
bash link_vrep.sh
For Windows, copy remoteApi.dll, vrep.py and vrepConst.py files from the V-REP installation folder to rl_car/vrep. To make it compatible with Python 3, replace the from vrepConst in vrep.py with from .vrepConst.
To run simulation, first start V-REP and load appropriate scene from rl_car/scenes. Both IR and proximity sensor examples contain a starter model written with Keras. Start simulation as,
# For infrared sensor
python -m rl_car.ir_sensor.train
# For proximity sensor
python -m rl_car.proximity_sensor.train