Reinforcement Learning

Adarsha Regmi
2 min readAug 1, 2021

The type of learning where agent explores the environment and learns how to perform the desired task. The agent carries out task exploitating the good outcome instead of bad outcome task.

amazon services

Training Reinforcement Learning model

It is the iterative process where the agent explores the environment continuously and collect experiences. The reward is provided for each step and thus the agent is able to finalize the good outcome in that reward basis.

Amazon deep racer is training a vehicle to run in a track and win the race.seems interesting right ? ?

The box shown above represent a step when a vehicle moves single step is called as individual state and each grid is marked as a number that is reward in this case. moving in the same line receives 2 reward whereas the more the distant grid the lesser the reward received.

scores for vehicle

The vehicle explores until it reaches goal or external grid with no reward. The single iteration is called as episode.

What to do in AWS ?

You are given sample reward function which you can write and update and evaluate accordingly.

Reward function can be updated and ran for different environment. Using deep racer I clarified my topics for reinforcement learning. Keep it trying and enter into league race…

trained model for 5 minutes
evaluation result

Thank you to Udacity Amazon Web Services