Prior to running training in the scenario, we should note the adjustment of a parameter that dramatically affects the computational cost. This is the horizontal sampling of the LDS, since the state of the robot is characterized by the set of range values in a given step of the simulation. In previous chapters, when we performed navigation in Gazebo, we used a sampling rate of 720 for LDS. This means that we have circumferential range measurements at 1º resolution.
For this example of reinforcement learning, we are reducing the sampling to 24, which means a range resolution of 15º. The positive aspect of this decision is that you reduce the state vector from 360 items to 24, which is a factor of 15. You may have guessed that this will make the simulation more computationally efficient. In contrast, you...
