This week’s community reading shifts attention from generated clips to the body’s evidence: touch, pressure, shape, and the small conventions that make an imitation feel faithful.

A vocabulary for contact

Atomic Motion proposes 13 signed translation, rotation, and hold atoms for each arm. A bounded spherical residual carries contact history without asking the controller to memorize an unbounded trajectory. In 7,520 offline interventions and 50 real trials per task, the authors report better separation of opposing actions and stronger out-of-distribution and force-responsive behavior. The useful idea is compositional: “push left,” “rotate inward,” and “hold” can be inspected independently before they are combined.

Feeling a whole hand from the wrist

Beyond Gestures uses a wrist-worn pressure array to estimate full hand pose and distributed contact force without electrical skin contact. The paper reports 4.6° mean finger-joint error in a single-user setting, and force R² of .57 from the band alone, rising to .75 when external pose is available. It is early evidence that a wearable need not observe every fingertip directly to recover meaningful contact structure.

Imitation without a borrowed body

SignMimic separates rigid canonicalization, non-rigid adaptation, and completion so a sign can be transferred across human shapes. Its promise is not a new avatar style but a cleaner distinction between the motion that carries meaning and the morphology that happens to perform it. The authors report state-of-the-art-level results on ASL50K, How2Sign, and CSL News; code and checkpoints are forthcoming.

These projects share a design principle: preserve the invariants that matter, and make the nuisance variation explicit. That is valuable for artists, accessibility researchers, and robotists alike. It also gives communities a better question than “does this look human?”—which parts of the interaction were actually preserved?

Readers may want to pair this with our recent motion-language roundup and the earlier sound and movement review. Both frame multimodality as a question of alignment; this week’s work adds contact and morphology to the alignment problem.

References