INS11-1566
Physical Problem Solving by Expressing Form and Property
Topic
S11. PMSE–PSK50 Anniversary Symposium: Advancing Polymer Science for a Sustainable and Intelligent Future
When and Where
Oct 1, 2026
10:45 - 11:10
Room 110
Session Chairs
Christopher SOLES
Christopher M STAFFORD
Presenter(s)
Jiyun Kim (UNIST)
Co-Author(s)
Abstract
As AI extends from language into the physical world, progress has come mostly along one axis - behavior, learned from ever-larger models and data. Yet what a policy can ultimately achieve is set by the body it runs on. The body is not a passive vessel but a design axis of its own, and a growing view across embodied intelligence and physical AI treats morphology and materials as active parts of how a system acts. We approach that axis from the materials side, programming structure and property so that a body's form and property become control variables set at runtime rather than fixed at fabrication. This path led us to mechanical pixelation: building a body from discrete material units whose arrangement, switchable at runtime, sets what it can do. We illustrate it with a few platforms - stiffness-tunable tensegrity that widens its configuration and behavioral space by bridging stabilized states; digital mechanical metamaterials that communicate through an LLM interface; and folding-based robots that express both shape and property patterns. Together they expand the space a body can reach without redesigning hardware, and because the units are discrete, that reach becomes something we can enumerate and compare.
But working across these platforms, a deeper question kept surfacing. Behavior now has a mature language - learned, bounded, mapped - yet it stops at the controller; the body has no such account, no language for what a given organization can do, and what it cannot. What is new is not optimization - searching this space for one good body - but a chance to understand the space itself. In that regard, today's AI lets us map, compress, and navigate the space a body’s organization opens. Our group therefore aims to study the physical capability written into a body - the space of what it can reach, its structure, and the boundary beyond - as an axis for meeting the world and solving physical problems.
But working across these platforms, a deeper question kept surfacing. Behavior now has a mature language - learned, bounded, mapped - yet it stops at the controller; the body has no such account, no language for what a given organization can do, and what it cannot. What is new is not optimization - searching this space for one good body - but a chance to understand the space itself. In that regard, today's AI lets us map, compress, and navigate the space a body’s organization opens. Our group therefore aims to study the physical capability written into a body - the space of what it can reach, its structure, and the boundary beyond - as an axis for meeting the world and solving physical problems.
