Defining Awesome

Haven’t read this yet but it doesn’t matter I’m thrilled. Two of my childhood Idols in one interview: link

“Man is something to be overcome”. This is what must happen if I make these bots. It’s tough. I am struggling on and off with them each day.

Not mine but good. Quote: “Almost all programming can be viewed as an exercise in caching.”

I was just shocked at how someone can optimize code. I used Justin Heyes-Jones (http://ping.fm/ui2zB) implementation of A* for pathfinding. It took 31s to calculate a route from one end of the map to another. Now I found this by someone called Konstantin Stupnik (http://ping.fm/TiUPL). He optimized Justins code by using heaps and hashing. I thought it would optimize the algorithm 2-3 times. For the same test from one end of the map to the other the calculation took 0.03ns ! I guess I don’t need anymore optimizations.

I can use my bots to find collision bugs. They will use them if its to their benefit: http://ping.fm/tttj2 They can’t crouch yet so to get to this place it invented a very complicated route using collision bugs: http://ping.fm/uQTWP

I’m brainstorming ideas on how to optimize the A* pathfinding algorithm. It looks fast but it will be more complex in the end and I will need it to run even every second for many bots at a time. I need some serious speedup ideas so if you have them please let me know.
Some ideas I’ve been considering:
– splitting the nodes into more general layers (this seems like the easiest approach but thinking about it I don’t see it as correct, how can it preserve information about the cost of the subroutes and shortest path information)
– precalculating the routes into memory (for all routes this will be 100s of MB’s unless there is a smarter way of calculating)
– using network theory (every network has hubs which are critical for the network, maybe utilize this?)
– use the GPU (?)
– calculate the routes iteratively as the game flows, instead of calculating it all at once (this is my best idea yet but it has a minus such that the bot will stand and think for a second before moving to a complicated position. But this is realistic behaviour cause it looks as if a human player is thinking and planning.)

First preview video of bots path-finding:

Has anybody ever encountered a resource/article about an AI trying to predict a physics based jump?

I decided to fix the rest of the test bugs later. Now I’m focusing on bringing ASAP a gameplay prototype. Meaning I’m working on bots again.

LD Build 1548:
– smoothened run and walk
– fixed getting stuck on cube after jump
– fixed walking in air
– option to disable crashrpt and autoupdate from cmdline
– limited normal mapped lights to 4 (do the ATi HD 4x cards work now?)
– key up can be used for climbing
– actors not loaded if map not created
– fixed climbing through wall in some parts of map