The Future of Roblox Avatars
Roblox has been making a flurry of announcements about its avatar system recently, and this week Anthony and I sat down to unpack all of them: the avatar joint upgrade, the dynamic head migration, the push towards AAA fidelity, and the growing role of AI in all of it.
Watch the full episode on YouTube.
Avatar Joint Upgrade: Ragdoll Physics and Beyond
The headline feature is the avatar joint upgrade, which introduces physics-based ragdoll behaviour using animation constraints and rigid constraints. Announced back in January as an opt-in Studio beta, it's now moving to opt-out for all existing experiences from May 2026. That means legacy games - including those whose creators left the platform years ago - will have their classic "fall to pieces" death animation replaced with ragdolling by default.
What makes this more than a visual tweak is the granularity. You're not just toggling between ragdoll and not-ragdoll; you can control the amount of force your animations supply, easing characters into a ragdoll state or having them fight back against gravity. Anthony called it "without a doubt the best physics upgrade we've ever seen to characters."
The upgrade also introduces support for bones within limbs. With the same 15 parts, you can now have multiple bones in a hand to control individual fingers, or articulate toes. The post even included a video of someone wiggling their fingers on their avatar - equal parts impressive and unsettling. Roblox emphasised that bones and extra articulation points are optional; the mandatory change is the underlying constraint system replacing Motor6Ds.
The R6 Question
One of the most debated aspects of this rollout is what happens to R6. Behind the scenes, R6 avatars are being rebuilt as R15, with a visual wrapper to preserve the classic look. Roblox has introduced an adaptive animation retargeting system: animate once for R15, and it automatically adapts to R6 and other avatar types.
The concern is whether "supported" really means supported when the underlying system is fundamentally different. Is the performance equivalent? Is the behaviour identical? We're not entirely sure, and there's a contradiction in Roblox's own messaging - phase two says Motor6Ds will still be supported for legacy content, but phase three says the avatar joint upgrade property will be removed entirely, forcing all places onto the new system.
For competitive shooters and obbies that rely on consistent hitboxes and predictable collisions, R6's simplicity has always been a feature, not a limitation. Those games will likely continue to use whatever Roblox calls R6, even if it's a wrapper around the new rig.
Dynamic Head Migration and AI Suspicions
On March 23rd, Roblox completed the full migration to dynamic heads. All legacy and classic heads and faces are now dynamic heads, with 17 distinct controls for facial expression. Static 2D faces are gone.
The migration hasn't been entirely smooth. Some faces look noticeably different in their default state - flat mouths, frowning expressions that don't match what players originally bought. MaximumADHD exported several migrated heads into Blender and found weight painting that doesn't look like it was done by a human: asymmetric blobs on otherwise symmetrical faces, oddly named layers, and patterns more consistent with automated or AI-generated output. Whether it's traditional automation or AI, the quality has been inconsistent enough to fuel suspicion - and for players already unhappy about the migration, it was the straw on the camel's back.
Why Is Roblox Doing All This?
All of these changes point in one direction: Roblox wants its platform to be indistinguishable from standalone AAA titles. They've been talking to creators about hiding the Roblox UI entirely, bringing in outside studios through the Incubator and Jumpstart programs, and pushing genres that are currently underrepresented on the platform.
When we spoke to Dave Baszucki earlier this year, he was clear that Roblox felt they had to expand - and that AI was a key tool to democratise the ability to create AAA-quality content. But with OpenAI shutting down Sora and the broader AI industry grappling with profitability and copyright, the question is whether AI-powered creation tools can actually deliver on that promise at scale.
Our take: AI will probably find its place in the tooling layer - avatar auto-setup, weight painting, rigging - where it's faster and more consistent than a human for tedious, well-defined tasks. The dream of AI generating entire real-time gaming experiences still feels a long way off.
Sources: