Ángel Martínez-Tenor - 2016
This repository provides a Reinforcement Learning framework in Python from the Machine Perception and Intelligent Robotics research group (MAPIR).
Reference: Towards a common implementation of reinforcement learning for multiple robotics tasks. Arxiv preprint ScienceDirect
Setup
- Create a python environment and install the requirements. e.g. using conda:
conda create -n rlrobot python=3.10
conda activate rlrobot
pip install -r requirements.txt
# tkinter: sudo apt install python-tk
Run
- Execute
python run_custom_exp.py(content below)
import exp
import rlrobot
exp.ENVIRONMENT_TYPE = "MODEL" # "VREP" for V-REP simulation
exp.TASK_ID = "wander_1k"
exp.FILE_MODEL = exp.TASK_ID + "_model"
exp.ALGORITHM = "TOSL"
exp.ACTION_STRATEGY = "QBIASSR"
exp.N_REPETITIONS = 1
exp.N_EPISODES = 1
exp.N_STEPS = 60 * 60
exp.DISPLAY_STEP = 500
rlrobot.run()
-
Full set of parameters available in
exp.py -
Tested on Ubuntu 14,16 ,18, 20 (64 bits)
Tested: V-REP PRO EDU V3.3.2 / V3.5.0
-
Use default values of
remoteApiConnections.txtportIndex1_port = 19997 portIndex1_debug = false portIndex1_syncSimTrigger = true -
Activate threaded rendering (recommended):
system/usrset.txt -> threadedRenderingDuringSimulation = 1
Recommended simulation settings for V-REP scenes:
- Simulation step time: 50 ms (default)
- Real-Time Simulation: Enabled
- Multiplication factor: 3.00 (required CPU >= i3 3110m)
Execute V-REP
(./vrep.sh on linux). File -> Open Scene -> <RL-ROBOT path>/vrep_scenes