Chapter 10 Training a Fetch Robot. Part 2. Robot does not seem to be learning

Course Support Using OpenAI With ROS
1

Question 1: The robot is moving but does not seem to be learning, almost always with negative rewards while running for the first 30 minutes and then shows “[ERROR] … Failed to fetch current robot state”. Did the robot learn when you guys are test running it and at how many steps and episode was this learning behavior observed? If not, how do you get it to learn?

Question 2: This line “state = self.env.reset()” restore the cube and robot to their original position but does not restore the robot arm to its original configuration / posture. Do you need to? If yes, how do you do it?

Thank you.

2

Question2: The Gazebo sims don’t take well well restarting simulations, most of the controllers crash . So the simulation starts and the arm is then placed in an init position each time

Question1: The training wasn’t very good and the notebooks is intended to teach you how to get started, but the training needs a lot of time and work.