Rapid pace multiplayer physics replication, are we there yet?

Community Community & Industry Discussion
,
1

Hey everyone. To be upfront on my knowledge base, I am an Unreal newbie. My realm has been modeling, messing about with metahumans/character animation through webcam, and of course a stupid amount of played games.

I’ve just been curious about the viability of legitimately good multiplayer physics for combat games. Smaller games like Half Sword and Exanima have shown the fun that can be had with the drunken fighting that is procedural animation style physics, and I know many people (myself very much included) have wished that multiplayer could be added to both of those games specifically. Having said that, my understanding is that implementing multiplayer replicability without gnarly amounts of desynch just isn’t really doable right now. for things like sword fighting and physics replication. Am I wrong in that assumption or is it just not something anyone has felt the need to pursue? Does the newest implementation of Chaos physics with 5.7.1 open up any possibilities?

I know I would pay stupid money to see a coop dungeon crawler with Fight Night Champion style controls mixed with the physics similar to something like Halfsword.

Hopefully someone with the ability to explain it to a layman like myself can shed some light on the topic. Thanks!

2

short answer: doable, but not with out-of-the-box replication. default replicated physics is server snapshots with no prediction, fine for a rolling crate, terrible for fast combat bodies, thats the desync people warn about.

what changed: async physics plus two prediction paths. resimulation (server rewinds and replays inputs, best for player-driven bodies) and predictive interpolation (client extrapolates props). party brawlers get away with partial authority because nobody cares about cheating there. full procedural fighters with every limb simulated are still mostly single player because cost scales with body count, not player count.

realistic starter path: keep the fighter kinematic (character movement has real client prediction) and simulate only a few key bodies like the weapon with resimulation. full ragdoll-every-limb multiplayer is a research project, not a first netcode project.

best mindset primer i know for what breaks and why: Networked Physics - Fundamentals
then the pawn build: Networked Physics - Pawn Tutorial