PFG

[IkerZamora]

• 13 semanas x 5 días a la semana = 65 días
• 5h al día x 65 días = 325h
• Methodology: Link1, Link2

[IkerZamora]

Day 1, Enviroment Setup & First Approaches to Reinforcement Learning (RL)

Setup the following software:

• Ubuntu 14.04
• Ros Indigo
• Gazebo 6.1.0
• Erle Plugins

First Approach to RL: Tutorial 1: Q Learning

• Read, understand and reproduce
• Read about RL here

[IkerZamora]

Day 2, Preliminary studies (Q Learning)

• Alpha VS Gamma VS Epsilon
• http://matpalm.com/blog/drivebot/
• http://www0.cs.ucl.ac.uk/staff/d.silver/web/Teaching.html
• Install kobuki + LOTS OF DEPENDENCIES (still dependencies left)

[IkerZamora]

Day 3, Preliminary studies & Testing kobuki

• http://www.cse.unsw.edu.au/~cs9417ml/RL1/introduction.html (intro + td learning)
• Setup kobuki + First launch + Keyop
• First ROS script in python to move kobuki
• Add lidar sensor to kobuki

[IkerZamora]

Day 4, Baseline of an Obstacle avoidance implementation

• Following the structure in drivebot, implement a baseline for the obstacle avoidance behaviour with kobuki in gazebo (turn left, turn right, move forward)

[IkerZamora]

Frameworks and Tools

• ROS Indigo
• Gazebo 7
• Gym by OpenAI
• Theano + Keras

Others

• Several ROS packages and libraries
• Python2.7 (?)
• APM (ardupilot, mavros, etc)

[IkerZamora]

Dissertation Examples

• https://addi.ehu.es/bitstream/10810/15880/1/GoiatzIrazabal_GAP.pdf
  • Nice project management
  • Basque
  • HW
• https://addi.ehu.es/bitstream/10810/15877/2/MikelBalducielDiaz.pdf
  • Nice structure
  • Spanish
  • SW
• https://addi.ehu.es/bitstream/10810/15885/1/Memoria_GAP_EnekoIrazusta.pdf
  • Overall good dissertation
  • Basque
  • SW
• https://addi.ehu.es/bitstream/10810/15891/1/GAP_EiderIrigoyen.pdf
  • Basque
  • SW
• https://addi.ehu.es/handle/10810/13356
  • English
  • SW
• Other dissertations: https://addi.ehu.es/handle/10810/13137

[IkerZamora]

https://www.inverse.com/article/21290-deep-reinforcement-learning-teaching-artificial-intelligence