Know when to replan

[lotu]

Instead of planning every move which wastes time the agent should only plan when it has a reason to. (Or in it's spare time if we get that option. The agent must detect when the state of the world has deviated and then replan around the new state.

[lotu]

In the re-planing branch we figure out where chip should be when he tries to apply each move and replan if he has deviated.

[darrenks]

This is how I think we should go keeping track of the state of the world and do replanning:

First we need a tool that can load the pddl state and domain and apply operators. I will look for libraries for this.

After calling fastdownward and receiving a plan. Start executing the plan, while also applying the operators in our own state.

When we get feedback from the screen, convert this to pddl and see if it matches our own state.
-if it does, then good
-if it doesn't, then update (overwrite) our own state (we learned something, or our model is wrong, or something non deterministic happened)
--if applying the existing plan with this new state still reaches the goal, then good
--if it doesn't then replan now