CODEWITHSUNDEEP
pythonpython-3.xnumpyreinforcement-learningq-learning
iuhettiarachchi
iuhettiarachchi

Reputation: 451

Q learning algorithm for robot where next state is not defined

I am new to machine learning and i developing a robot which environment is dynamic. I am using python as the programming language for my project.

I have a goal state and robot has four actions such as forward, backward, turn right and turn left.

The problem is I cannot define the next state when robot take a one action. But I can identify state is goal state or not after each episode.

How can i apply Reinforcement learning model for this. Can i use pomdp ???

Upvotes: 0

Views: 268

Answers (1)

shunyo
shunyo

Reputation: 1307

Even if the environment is dynamic, you should have some notion of motion model where the discrete action should lead to the robot advancing "one step" straight. In a gridded world, it is easy, however, in a continuous environment, you can grid the world up and still use discrete action. If the next state is undefined, no machine learning/decision making algorithm can work.

Upvotes: 1

Related Questions