Experiment
vr-swords
activeA PCVR sword fighting game where opponents are "ghosts" replaying recorded human motion. Records 60Hz pose data, tags outcomes, evolves a population of sequences via Elo and tournament selection, and serves the strongest as the next opponent.
What happens when the “AI opponent” isn’t a model at all, but a search-and-evolution system running over recorded human motion? The generation happens in a headless combat sim that runs thousands of duels between stored pose sequences, and the winners get played back at you in the headset.
Stuff it’s surfacing:
- Embeddings as a search index for physical motion: DTW plus small embedding filters to pull up “what would a person do here” without training a policy
- Elo as a fitness function for something you can only feel, not measure: does this swing land like a real fight
- Headless evolutionary loops running overnight to fill a content pool nobody had to author by hand
Field notes
No posts tag this experiment yet. When a post does, add experiments: [vr-swords] to its frontmatter.