foot contact correction in mocap — anyone found a non-terrible workflow?

something that's helped me a lot before touching any IK pinning: doing a tangent cleanup pass in the graph editor first. mocap data imported into blender tends to have auto-tangent handles with tiny overshoot artifacts around keyframes. nothing visible on its own, but when you layer IK constraints on top, the constraint ends up fighting these micro-bounces and the foot looks slightly floaty even at full constraint weight.

flatten the tangents specifically on the frames you've identified as contacts (switch to linear or constant just for those keys) before setting up any pinning. takes maybe 10 minutes on a full session and the constraint behavior gets noticeably cleaner. the foot actually stays where you put it instead of drifting a few millimeters over the next two frames and making your whole pin feel soft.

One thing that really improved my cleanup workflow: treating contact identification as its own completely separate pass, disconnected from any IK work. Before touching constraints, I export the vertical position curve for each foot joint and find the local minima. Those are your candidate contact windows. Then I verify the foot is also moving slowly horizontally during those same frames to confirm it's actually planted and not just at the bottom of a step cycle.

The result is a frame-accurate contact map that answers "when is this foot down" before I open MotionBuilder or touch a single IK key. The actual pinning pass becomes almost mechanical because all the judgment calls are already resolved. I do this in a spreadsheet, which sounds archaic, but it means I'm never making decisions and doing work simultaneously, just executing against a plan.

The workflow that actually helped me with Rokoko foot sliding was doing the contact pinning pass in MotionBuilder before taking the data anywhere else. Set up IK effectors on the feet, keyframe-pin them at the contact frames, and let the solver handle the leg geometry. It's not perfect, but the residual cleanup afterward is way smaller. You're fixing small oscillations instead of whole-leg drift. Doing that same work manually in Blender's graph editor is rough because correcting one rotation value cascades through the whole leg chain and you end up chasing it forever.

Also worth double-checking: make sure you're exporting at the session frame rate. I had an entire session I thought had foot slide that was actually a 60fps-to-30fps bake interpolating straight through contact frames. Lost two hours to that one.

Something that helped me a lot before touching any IK pinning: doing a velocity analysis pass first. Compute the world-space velocity of each foot bone frame-by-frame and mark where it drops close to zero. Those are your actual contact frames, and they're usually more reliable than trying to identify them visually on the curve editor.

In Blender you can script this fairly quickly. Iterate the action's FCurves for the foot location channels, compute finite differences between adjacent sample values, and either write the results to a text block or use them to insert markers directly on the timeline. The velocity data also makes it obvious when the capture itself had a problem. You'll see frames where foot velocity spikes mid-contact, and no amount of IK pinning will fix that cleanly. Better to know before you spend an hour on it.

Has anyone tried routing mocap data through Cascadeur specifically for foot contact cleanup before taking it anywhere else in the pipeline? I started doing this about six months ago and it's become permanent. The physics-based foot pinning tools feel more interactive than anything I've used in MotionBuilder for this specific task, and since it imports and exports FBX cleanly, it doesn't require major pipeline surgery to add.

The AutoPhysics pass is hit or miss on complex motion, but the manual contact pinning tools are fast once you know them. Curious if others have tested it with Rokoko data specifically. My sessions have been coming in pretty clean, so I'd be interested to hear how it holds up on noisier captures.

One thing that sped up foot cleanup a lot for me: before touching any IK constraints, do a dedicated pass in the graph editor just to identify and mark the frames where each foot should be fully planted. Select the foot bone, look at the world-space Z translation curve, and find the local minima that correspond to actual contact. Mark those frame ranges explicitly before you do anything else.

Once you know exactly which frames are contact frames, all the correction work has a clear target. You're not scrubbing through the viewport trying to feel where the slide begins — you have specific ranges to work within. IK pinning or manual curve flattening is much more targeted that way. Doesn't remove the work, but removes a lot of the guessing about where to start.