Synopsis

Machine learning Praktikum 2016 at FZI Karlsruhe
Reinforcement Learning with Spiking Neural Networks
Group 2: Daniel Geier, Simon Di Stefano, Fabian Mack, Lea Steffen

Getting started

To start on specific lanelet:
1. run restart_'worldName'.sh (Starts Gazebo)
2. run camera.py (Gets Camera Sensor Image, and publishes the preprocessed image)
3. run carcontrol.py (Outside lane controller)
4. run neuralnet.py (everything else)

Inventory

./src/camera.py
Gets image from camera and preprocesses the image (retina)

./src/carcontrol.py
Sets car back on track

./src/cockpit.py
Plots net structure including weights.
Plots speed, distance, reward, angle.
Offers functionality for debugging

./src/neuralnet.py
Builds desired net structure, includes learning algorithm, driving behaviour and world state
Main classes: BaseNetwork, BraitenbergNetwork, DeepNetwork, Learner
Parameter (plotting & logging): main(argv) n.plot, n.log

Custom roads: ./worlds/onecrossing.world, ./worlds/tcrossing.world
Lanelet specifics: ./worlds/lanelet_information.cpp, ./worlds/lanelet_random_pos.cpp