Creating an OpenAI gym environment as a wrapper for Gazebo robots simulator, so that the environment would be a bridge to connect an agent as a brain to the robot's body and interact with the Gazebo world, is the primary goal of this project. In addition, we will try to use Deep Reinforcement Learning methods to implement DQN agent as baseline and use a robot to learn how to navigate in an unknown map, using sensory data as input to the network, and generating velocity commands for the robot as the output of the network. Autonomous navigation and obstacle avoidance, can be used for autonomous driving in an unknown environment that we don't have any map from that. It is also useful for dynamic environments in which the map is constantly changing. This project could be used for autonomous exploring and mapping the environments, providing another solution replacing well-known SLAM algorithm.

Student

Sepehr

Mentors

  • Alberto Martín
  • Francisco Rivas
close

2017