Wednesday, June 24, 2020


Hello everyone, it's update time again!

There's nothing here about glasses, the topic just sort of popped into my head because I had an eye exam and got a new pair yesterday. It's definitely an interesting experience when you're wearing a mask and the exam equipment keeps fogging up. It doesn't help that Guam is a balmy place at the best of times.

Work has been progressing as expected, and it's rather nice that nothing crazy seems to have happened so far this month. *knocks on wood manically* That being the case, I think it's a good time to talk about a decision we made regarding the NPC AI system. I've written a lot about the system, the plans, and how it works so I don't want to spend a lot of time rehashing that. You can find out more about it if you'd like in this blog, and there is also some information available in the game in the encyclopedia and AI training page.

The main issue is that while the system worked, and it was pretty cool to say we used learning software to have our NPCs make decisions... It hasn't turned out to be an optimal solution. One issue is the time it takes for the AI system to return a decision, which takes a couple seconds. It doesn't seem like much, but the pause is somewhat disruptive. The other issue is training. We've been using a simplistic neuron model and a fairly straight-forward set of training data, but we'd need to spend a lot more effort to get more training data and use a more complex model, to get a better set of answers.

So far, all we've managed is some vaguely useful output, and I suspect that it would take a lot of work before we could get the AI to the point where it is useable for more than a vague yes/no evaluation for date activities that frankly doesn't work very well. In the meantime, we've created more traditional AI setups for specific situations, because it has been easier to do that than get the learning software setup working well enough for it.

In any case, the decision was to stop using the AI system altogether and continue with more traditional methods. This means dating and hangouts that work more reliably and realistically in terms of how NPCs enjoy it. I had already made plans for this possibility, so it really doesn't require much effort to drop the old system. Learning software was a bit of an experiment, after all.

For now, we want to focus on making a game that works well. Having to write new traditional AI setups for NPCs in regard to relationships has highlighted the issues with dating that existed because of the poor learning AI. So we'll have a better dating experience with the upcoming release by replacing it.

That said, we aren't going to delete the work we've done on the learning software setup. We'll leave the learning-software-based AI in the game's source, and if we are able to spend time on it in the future to get it working at a reasonable level we'll be able to add it back in. I don't think that's very likely, to be honest, but I don't want to rule it out.

So, long-story-short, look forward to some better dating along with the new stuff in our next release :D

That's it for now, so until next time, thanks for your support!


No comments:

Post a Comment

The Combat Release - v0.36 Change Log

Hello everyone, it's finally time for the version 0.36 release! We've got a bunch of new stuff for you in this release, a mixture of...