Three projects released from 5–6 October 2026 expand embodied AI beyond a screen-and-prompt exchange. A remote hugging suit turns pressure into a communication channel, a clay-based reflection system lets generated interpretations remain negotiable, and an open-ended virtual body raises a harder question: when does exploration start to resemble play?

The Hugging Suit treats context as part of haptics

Hugging Suit, released 6 October, combines a programmable pneumatic actuator matrix, a portable high-pressure unit, and fabric pressure sensors to simulate remote hugs. The Research through Design process considered not only tactile resolution and wearability but also lighting, privacy, and whether the remote partner was visible by video.

The authors describe preliminary feedback suggesting greater comfort and emotional engagement in low-lit private spaces with visual contact. That does not establish a universal emotional effect. It does show why pressure amplitude alone is an incomplete account of mediated touch: the same squeeze enters a different experience depending on setting, consent, expectation, and relationship.

ClayFlect keeps interpretation in the participant’s hands

ClayFlect, also released 6 October, combines tactile clay-making with conversational and visual generative AI. In a mixed-methods study with 50 participants, reflection moved among the material object, conversation, and generated representation. Participants compared, redirected, reinterpreted, selectively used, or ignored what the AI produced.

The clay mattered because it remained directly manipulable without AI. Generated images could externalise possibilities, but they did not become the final authority on what an artefact meant. The system’s design implication is unusually useful: preserve parallel sites of control and allow AI involvement to deepen or recede.

A virtual embodied agent begins activities without a task

Is this machine playing?, released 5 October, placed an AI coding assistant in an unfamiliar virtual body with no specific task or reward. Across 30-hour runs, agents climbed, stacked blocks, drew patterns, tested physics, and developed distinct histories. Nathan Cloos and colleagues compare these recurring behaviours with classical criteria for play.

The title is a question, not a settled conclusion. Repeated open-ended activity may resemble play behaviour without establishing subjective enjoyment or intention. The productive community question is methodological: what observations would distinguish exploration caused by the system setup from play as a mode of development?

The common thread is negotiable agency

These projects do not merely add another modality. They redistribute who can redirect the interaction. A remote partner modulates pressure, a clay-maker can ignore a generated interpretation, and an agent without a prescribed goal explores possible activities. That connects with Somatic-AI Lab’s account of shared agency in dance cyphers and Lian Loke’s emphasis on designing from the lived body outward.

For practitioners, the review question is simple: can a participant pause, reinterpret, or continue without the AI? If not, “collaboration” may describe the interface while control remains one-sided.

References

  • Hugging Suit: Pneumatically-actuated system design for remote haptic experiences. (2026). arXiv. https://arxiv.org/abs/2610.08305
  • Sim, K. Y. H., Nguyen, Q.-N., Han, S. Y., & Choo, K. T. W. (2026). “I’m very happy for it to start hallucinating a little bit”: Using ClayFlect to negotiate multimodal AI representations in material meaning-making. arXiv. https://arxiv.org/abs/2610.08943
  • Cloos, N., Norelli, A., Durbin, D., Andreas, J., Rus, D., & Isola, P. (2026). Is this machine playing? arXiv. https://arxiv.org/abs/2610.07130