goap | Simple Goal Oriented Action Planning demo | Game Engine library

I came up with a relatively simple solution. I ended up using the command line player mpv with the —no-video option and the YouTube URL. I can confirm the audio only gets decoded.

Well, the solution ended up being pretty simple. Here's the correct way to write the command (I was missing a '=' after the --sout option and didn't need the quotes around the parameters; this is different in linux):

Nodes are indeed WorldStates. And Edges are actions. But note that they are directed edges!

Where the pre-conditions come in: They determine which edges (actions) flow out of the node. Only actions that have their pre conditions met, are valid edges exiting that state node.

So for finding the neighbours of a node, you would check each action if all pre-conditions are met. If so, apply the postconditions to see the node that the action will lead to. The action is then a valid edge between these states (nodes.)

Please see the open sourced GPGOAP (General Purpose Goal Oriented Action Planning) for an implementation of GOAP with A*. It's straightforward code in C that explain all the steps. I am the author of GPGOAP.

Thoughts on regression

Now for the regressive part: I've never implemented a backward search from goal to current world state. So I can be of limited help on that front.

Two neighbouring nodes will still be connected on the basis of one single action. You now enable/disable edges not based on precondition of the action, but the post-condition of the action. If the post condition does not match the current node, then the action will not be valid. If it does, I expect you add the neighbour by forcing the pre-condition of the action.

What's your reason to prefer backwards search over forward search?

The comparator function of a sort should return -1, 0, or 1. Returning a boolean instead (effectively 1 or 0) will fail to sort the list correctly!

For example, in Chrome (sort implementations are host-dependent), this line: