Predicting Faster Than We Can Make: Polymer-Specific Process Physics as a Frontier for Autonomous Soft-Matter Discovery.
Machine learning is now a productive tool in polymer research, predicting many properties and proposing new structures by generative design. For the materials whose behaviour is set by processing rather than chemistry alone, our capacity to propose polymers is beginning to outrun our capacity to make, process, and validate them. This Perspective argues that a key bottleneck for these materials is not model architecture but a polymer-specific gap: macroscopic, processing-dependent properties and slow validation timescales. I distinguish chemistry-dominated, condition-dependent, and process-history-dependent properties, and develop two contrasting cases: sustainable thermoplastics, governed by melt flow and thermal history, and supramolecular hydrogels, governed by aqueous self-assembly. I propose and stress-test a self-driving laboratory built around the parts the autonomous-materials literature has under-served for polymers: handling viscous, non-Newtonian melts and sol-gel transitions; preserving shear and thermal history; surrogate simulators to keep the loop fast; and mechanism-preserving accelerated tests for slow properties. The contribution is not the closed loop, which is well established, but the claim that, for non-dilute melts, concentrated solutions, and assembled soft matter, and for properties whose validation runs to months or years, polymer process physics and slow-property validation are among its principal frontiers. Investing there turns computational acceleration into deployed materials.