Hands-On ROS for Robotics Programming

Reward table

This table specifies what reward the agent gets for every action it takes. A well-designed policy incentivizes the most desired actions with greater rewards. For the case of the cab example, the reward table is as follows:

• The agent receives +20 points for a successful drop-off.
• It loses 1 point for every time step. This way, we encourage it to solve the environment as quickly as possible: all the time it is on the road it is consuming resources, such as fuel, so this negative reward can be understood as the fuel expense.
• It is given a 10-point penalty for every illegal action it performs (during the pick-up or drop-off actions).

Next, we proceed to describe the action and state spaces that the cab has to comply with to evolve ...