51:00 Concrete Example of Hector's language based representation of policy 57:50 Concrete Example of Hector's Langague based representation of Sub-goal structures (intrinsic rewards)
### Prerequisite : - Have basic understanding or FOL(take quick look on any website) - STRIPS www.sci.brooklyn.cuny.edu/~parsons/courses/3410-fall-2012/notes/lect07.pdf ### Atom A Minimal Non-Decomposable Fact . like "Mercy is a Cat" . also written as Cat(mercy). So, "Mercy is a Cat who likes butter" || Likes( Cat(mercy), Butter ). is not an atom. ### Classical Planning You have set of facts that describes the current state of world : Block A is on the table || OnTable(A) Similarly, OnTable(B) and OnTable(C) Suppose Goal State is : A, B ,C stacked on top of each other. And represented as : > OnTable(A), > OnTop(B,A) | B is on top of A > OnTop(C,B) Suppose there's a The Robot can do actions A like : 1. PutOnTop(x,y) >> Put x on top of y 2. Carrying(x) >> Carry object x CLASSICAL PLANNING ALGORITHMS are basically SEARCH algorithms that finds a "sequence of actions" that takes Current State of World to Goal State. IW is one such algorithm. ### Width of Problem For a given problem, the minimal number of actions that's required to reach from current state goal state is called WIDTH OF PROBLEM.
Great video giving a high-level overview of a method of solving the Watchman Route Problem. Only thing I regret they didn't cover was a method to determine the optimal (at least somewhat optimal) starting route, which would allow a circuitous path if they used full TSP instead of the variation.
IKR. I really find it odd that for an efficient algorithm and suggested a lot in forums, it has really few resources for it. Aside the main paper, I can only find the introduction (which kept repeating things) by other people and one visualizer. I can't find a reliable pseudocode that I can understand. Nor a proper tutorial, just introductions.
if you have some questions, I did an implementation of the classic jps and the improved one.The hard part was the twikling bits i had to figure out. (and probably not as bright as what they did)
@Very_good Sorry I didn't get any notification of your reply. It's fine I would like to see d*-lite. I don't know why my reply is not showing probably the same reason why I missed yours I'm trying to write as new comment so hopefully you can see it.
@@daveoconnel8583 Absolutely will my man. It's written in C, I've exported it as shared object to use in Python. The implementation is not pure MPAA*, but rather D*-lite using the MPAA* optimization. Before I share I'd like to fix a few things if you don't mind. If you give me an email I'll send it over.
Conditional plans can be implemented as linked list. The IPL-V programming language was invented for the purpose of List processing. This is equal to self-modifying code.
