Skip to content

Mobile robot control in V-REP using Deep Reinforcement Learning Algorithms.

License

Notifications You must be signed in to change notification settings

Eriwdz/mobile_robot_rl

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

63 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Mobile robot control using Deep Reinforcement Learning

Short description

Following project was a part of my master thesis. Current version is a little bit modified and much improved. The project is using modified algorithm Deterministic Policy Gradient (Lillicrap et al.arXiv:1509.02971) (written in Tensorflow) to control mobile robot. The main idea was to learn mobile robot navigate to goal and also avoid obstacles. For obstacles avoidance, robot is using 5 ultrasonic sensors. For navigation task, required information (like absolute pose) are taken from simulation engine. That's a little hack. However, more realistic (like odometry) pose estimation can be found in gym-vrep.

How to use

Dependencies

Simulation

Project is using robotic simulation V-REP EDU. To setup V-REP environment follow instruction described in gym-vrep repository.

Python

All Python dependencies are in requirements.txt. Just run command:

pip install -r requirements.txt

Running

To run training just run command:

python main.py --train

To run testing just run command:

python main.py

For more options (like state, action normalization) run command:

python main.py --help

Result

Agent was learned for 1000 episodes and results are shown below:

alt text alt text

About

Mobile robot control in V-REP using Deep Reinforcement Learning Algorithms.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%