/ai_controlled_car

Group of nodes in ROS to control a model of a car using deep neural networks

Primary LanguageCMake

ai_controlled_car

This repository is meant to control a car model, using neural networks in order to follow a trajectory within his lane and avoid moving obstacles.

Watch it in action in the following video

car_working_image

Network architecture

This network was inspired by the residual neural network architecture, specially the densenet approach to concatenate the outputs of the skip layer connections instead of adding them, using just 48,300 parameters. model architecture

Tested with the following settings

  • ROS (melodic, 1.14.3)
  • Ubuntu (16.04.5 LTS)
  • python-catkin-tools

How to build

On the root directory (where this README.md file can be found) execute the following commands:

catkin build
source devel/setup.bash

How to use

Currently the simulation comes from another project, so please follow the instructions on the the next repository's wiki in order to set the car model simulator.

Then open a terminal on the root directory and execute the following commands:

roslaunch ai_control.launch