Add continuous maze example.

[mohsen-ghaffari1992]

@wasowski
Please consider reviewing this example and let me know whether it is what we want.

[wasowski]

Should isFinal be defined on MazeState or on MazeObservableState?

[mohsen-ghaffari1992]

According to the agent class, it should be maze state.

[wasowski]

Then, I think the code is incorrect either here or there. If isFinal is called on observable state, then the set of the final states should be an entire square, not just two points. If isFinal is called by symsim on observable state, then they should require the observable state type, not the system state type? Either side is confused. It does not make sense to check an observable state condition on a continuous state ...

[mohsen-ghaffari1992]

Then, I think the code is incorrect either here or there. If isFinal is called on observable state, then the set of the final states should be an entire square, not just two points. If isFinal is called by symsim on observable state, then they should require the observable state type, not the system state type? Either side is confused. It does not make sense to check an observable state condition on a continuous state ...

Maybe I wasn't clear, I meant it is defined on MazeState. If it was defined on MazeObservableState, you were right.

[wasowski]

In the continuous, version of the problem it would be more natural to change the distance moved, not the direction. But anyhow, how are you going to treat randomization in the symbolic executor?

[mohsen-ghaffari1992]

I have an idea that we do not need to do anything about randomization. Let's talk about it in detail in the meeting. Just as a short answer, when there is randomization, it means that instead of a1, the behavior of a2 must be considered. Since we are analyzing the program for all actions, then we should not be worried about randomized actions.

[wasowski]

But here a good randomization would not select a different action, but would result in adding random noise (for instance Gaussian) to the distance travelled.

[wasowski]

I think the package name should be continuousMaze not continuousmaze.

[mohsen-ghaffari1992]

I tried to keep it similar to the other packages. e.g. simplemaze, simplebandit, ...
Should I rename it?

[wasowski]

I have left my comments, but I am confused why tests for cartpole are failing. Have we added cartpole to main? (I might have simply forgotten...)

[mohsen-ghaffari1992]

@wasowski
Thanks for comments!
I edited the code based on your comments and push the new version.
Regarding to cart pole, yes, we have added it in the main.

[wasowski]

BTW. Andreas tells me that this example is basically the same thing that is known as the random walk example.

[mohsen-ghaffari1992]

BTW. Andreas tells me that this example is basically the same thing that is known as the random walk example.

Is it?
Random walk is referred to simple maze type of problems according to my knowledge. I googled it a little bit and I still think my understanding is correct.