My own implementation of Unity ml-agents. In this example the agent is a balancing robot, consisting of a body connected to two wheels by a revolute joint. The wheels are torque controlled. The goal is to move to the target (red line) without tipping over. The red line spawns randomly with an x-distance from the starting point. The agent is penalized for tipping over (-1), spending time (-0.1 per iteration), and awarded for moving forward (0.1 per iteration) and reaching the goal (1).