Deliverables for wednesday:
1. Submit a maze (based on the simulator specifications)
2. Come up with a player that does something interesting (Moving around, updating data structures etc).
The professor started of with a brief description of the original Adventure game of Crowther and woods, which provided the inspiration for this project.
[Class discussion about the project specification]
Group-1: Danny, Sean and Stuart.
Group-2: Jon, Green and Laima
Group-3: Shilpa, Colin and Nipun
Group-4: Smriti, Ben and Yang
Group-5: Sharadh, Manuel and Monal
What would you do if you had just one object?
Manuel: Drop it as soon as possible.
Ben: Drop it in the start cell.
Jon: Establish the path into the start cell.
Green: Advocates dropping it in the start room.
Yang: Drop the object second room, since it gives you more information as its outside the scope of the exploration.
Ben: If you are exploring, its always random. You have no information unless you id both the ends.
Manuel: If you establish a cycle between 2 rooms you have some concrete information about the maze.
Colin: Start discovering lots of cycles – in order to avoid them in the future.
Can you merge the information about cycles?
Smriti: You have more information from cycles about the rooms that do not lead to the treasure room. (A way to eliminate doors)
Laima: If you drop objects in a sequence, then you can – as you traversing the room – start putting in label on the passages.
You can get some locality information by eliminating all possibilities in terms of passages from a particular room.
Sharadh: The length of the cycle is not going to be proportional to the actual number of steps that you have taken to get there.
Jon: There is a high chance of staying at the same place if you have a sparse graph.
What kind of data structures do you need?
Nipun: Its important to have both the history of the graph and the passages. It depends on the number of objects that you have at your disposal.
Do we need to fully reconstruct the graph to do well on the game?
Laima: If you dont have good information about the graph you will spend more time just going round and round without any gain.
How is the mapping going to slow you down? Can you come up with a graph where random exploration is much cheaper than systematic exploration?
Smriti: Use a strategy that uses pseudo-random exploration for the map.
Sean: You have no information to say that there are infact in self loops.
Shilpa: If there are no cycles then random will be advantageous over systematic.
Sharadh: Bigger cycles are just as much a problem as smaller cycles.
Colin: You can minimize the number of cycles in the first 10 turns, and then use that information later.
Green: Have some objects in reserve so that in case you discover a cycle you can use the additional objects to fully explore that cycle.
Sharadh: But you may not be able to collate the information until you get back to the cycle. So there is a trade-off or balance involved in making the decision about the number of objects in reserve.
Nipun: Explore a local area with all your objects, and once you are done – pick all of them up and move on to a new area.
There are infact many different trade-offs involved.
Ben: You could be many moves away but you may still be in the same locality of the maze.