LIDAR based Obstacle Avoidance with Reinforcement Learning

This project lists all of the deliverables for the TUM project course Applied Reinforcement Learning (Summer Semester 2019).

Results

Reward Graphs:

State Representation	Linear Value Function Approximation	Algorithms

More graphs.

Project Demo.:

Simulation	Real Turtlebot

Supplementary Material:

• Sensor Model Notebook
• Auto-Encoders Notebook
• Source Code

Requirements

• Python 2.7 (Python 3 for sensor-model fitting and auto-encoder training)
• ROS-Kinetic with turtlebot
• Catkin
• PyTroch
• Scipy
• PyYAML

Instructions

1. Move the rl_tb_lidar and stage_ros_u folders to catkin_ws/src directory.
2. run catkin_make in the catkin_ws directory.
3. Run source devel/setup.bash command in the catkin_ws directory.
4. Run roslaunch rl_tb_lidar tb_stage_m1.launch to launch only stage.
5. Open an another terminal, go to the directory of the python script e.g. cd ~/catkin_ws/src/rl_tb_lidar/src and run python main.py configs/config.yaml.
6. To try different configurations, edit the configs/config.yaml file accordingly.

Versioning

We version the project with each new deliverable. For the versions available, see the tags on this repository.

Authors

• Akbar, Uzair - uzair.akbar@tum.de
• Gundogan, Alperen - ga53keb@mytum.de
• Ellouze, Rachid - ga63nix@mytum.de

See also the list of contributors.

Acknowledgements

• Dr. Hao Shen
• Martin Gottwald