Breaking the Physics of AI MoCap

A few days ago, I wrote about why I locked myself in a room to build Mimic—a native Android motion capture engine designed to give indie filmmakers their economic sovereignty back. No $2,000 suits, no corporate subscriptions, just a smartphone.

Today, I want to talk about how my team at ShardLabWorks just cracked the most notorious, unforgiving problem in the AI animation space.If you’ve ever tried budget or AI motion capture, you know exactly what I’m talking about: The "Skating." You import the raw data into your 3D software, and your character looks like they are sliding on ice. Their knees shiver, their ankles roll unnaturally, and the sheer amount of manual cleanup required makes the data virtually useless.

Most tech companies accept this as the cost of doing AI tracking. They dump the raw point-cloud data on the animator and say, "Good luck."As an indie filmmaker who actually has to edit these shots, that wasn't good enough for me. We realized we couldn't just rely on computer vision; we had to inject raw physics into the pipeline.So, we completely rewrote how the engine handles the lower body.

We built a custom "Anti-Skate" engine directly into the Android pipeline. Instead of just guessing where your foot is in space, our algorithms calculate Euclidean velocity and floor-plane intersection. When your foot hits the virtual floor, the engine physically locks it. It stays planted. No more sliding.But we didn't stop there. We tackled "Flamingo Knees."

Cameras suffer from depth ambiguity. When you lift your leg straight forward, the AI often thinks your leg just magically shrunk, causing the knee to glitch. We wrote a geometrical Inverse Kinematics (IK) solver that forces the knee to respect the anatomical radius of the human femur and tibia.

If the hip and ankle move closer together, the math forces the knee to bend outward naturally.

Finally, the workflow.

A hyper-optimized mobile app means nothing if the data gets trapped there. We wanted a studio-grade pipeline. So alongside the app, we built a native Blender plugin. You drop our .mimic file into Blender, and with one click, our plugin parses the physics data and flawlessly retargets it to a Mixamo or Unreal Engine 5 skeleton.This isn't just an app anymore; it’s an end-to-end studio in your pocket.

We are taking the heavy, expensive infrastructure of Hollywood mocap and compressing it into a free, ad-supported toolkit for anyone with a story to tell.We are so close to the finish line, and the stress-testing is yielding incredible results.I’d love to hear from the 3D artists, game devs, and animators out there: How much time do you currently spend cleaning up bad mocap data? What is your dream feature for a mobile mocap pipeline? Let's talk in the comments! 👇