?
Transfer of learned exploration strategies of a mobile robot from a simulated to real environments
Ch. OS4-4. P. 120–123.
Reinforcement learning based approaches show promises in various robotic applications, but a significant amount of time and resources are required for a robot to learn optimal behavior. Using virtual environments, we could significantly speed up and improve performance of a target task. We implemented a reinforcement learning based exploration algorithm for a mobile robot, training in Gazebo environment and transferring learned strategy to a real robot. We show that it is convenient and appropriate to use simulation to train strategies for mobile robots.
In book
ALife Robotics Corporation Ltd., 2019.