Robot Arm Environment for Reinforcement Learning

Multi arms setup for training an Reinforcement Learning algorithm.

Install

git clone https://github.com/MorvanZhou/RLarm
cd RLarm 
pip3 install -r requirments.txt

Training

state: distance to goal for each join, angle to goal for each join, if touches goal All distances to goal are normalized to a range of [0, 1]. All angles to goal are normalized to a range of [-1, 1]. 1 (touches) / -1 (not touches) is the last state.

State example: array([ 0.04573185, 0.13328803, -0.03580408, 0.47065321, -1. ])

Train command:

# train with 3 arms
python3 main.py -n 3

# train with 2 arms and hide window
python3 main.py -n 2 --hide 

-n: number of arms

Testing

After training, run following command to test the last stored model.

# load 3 arms on 340 epoch
python3 main.py -n 3 --load 340