Show HN: Forever-Time Multi-player- Play with your loved ones after you're gone
omeys.itch.ioSummer Friends Don't Stick Around is a "forever-time" multiplayer game created to immortalize players and give others a chance to play with them "in spirit" even after they are no longer with us.
In the "Remember Me" mode, players can train a neural network to capture their play style. The output is a data model that can be shared with friends and family. Your playstyle is essentially encoded in the data model.
In the "Remember Them" mode, players can load a data model file and play with them.
I'm still learning more about neural networks and tweaking them to capture more of how any particular player approaches the game. There is room for improvement. This is an interesting use of AI I hadn’t thought about before, but now I can see a number of use cases for technology like this. Can you share more technical details about your approach? (It’s cool if you can’t.) So, I had this idea from a heartfelt meme that I saw online many years ago about someone who saved their XBOX because it had a save file from their dead brother. The save file was a replay of their last run around a particular track in the racing game. I was really moved by the meme and wanted to create a game that sort of "immortalizes" you like that. When I learned about machine learning, I identified it as a way to make that happen.
To make the current game, I started with a car driving around a track. The input data for the neural network was the distance from the central spline, tangent angles, and the preceding and the following curvature of the spline. It worked well but I wanted to add different gameplay than just driving around a track.
In the current iteration, the players are chasing a rabbit around the park and the input data is much simpler than in the first iteration - just the distance and angles. I'm only collecting 15000 data points right now, 5 minutes of game time at Unity's 50 physics per second.
I originally wanted the player to be able to play alongside multiple other data models that may choose to do different activities in the park. So, there was a track, a rabbit, and a butterfly - for each target, I collected different data points. But I wasn't able to make a network that accurately determined what they originally chose to do - i.e. which target they preferred - drive around the track or chase the rabbit or butterfly. The goal was to truly capture what a player preferred to do in the game space rather than just how effectively they were able to chase one goal. I'm still working on that. This is what Drivatar tech does in the Forza games on Xbox, it learns your playstyle so you can be part of other players sessions while you're offline. Well I feel foolish now not knowing this existed but it was fun while it lasted. If someone else thought it was worth doing that just validates your idea. Who cares? Build what you want. I liked your idea. I must say I'm very intrigued by this idea, honestly not so much for playing against an AI simulacra of an old friend, but for training up AI to play games without the traditional method of "give them cheats to make them harder!" So similar but not the same - I’ve been researching how to build my own chatgtp as a service - called something like ‘whenImGone’ - so a user can train a bot to talk like them, with a really sophisticated questionnaire you can pump it full of info so your family and friends can talk to it, at important moments, birthdays, anniversaries, to comfort them kind of like a game when your dead. Inspired by the black mirror episode - remember me. Want to team up? Hi, that is interesting to me. I found your GitHub profile but can't find an email. I'm actually interested in other usages of such technology but this is interesting.