From smart self-tightening sutures and expandable stents to morphing airplane wings, shape memory structures are increasingly present in our daily life. The lack of methods for synthesizing intricate structures from them on the micron and submicron level, however, is stopping the field from developing. In particular, the methods for the synthesis of shape memory polymers (SMPs) and structures at this scale and the effect of new geometries remain unexplored. Here, we describe the synthesis of shape memory polyurethane (PU) capsules accomplished by interfacial polymerization of emulsified droplets. The emulsified droplets contain the monomers for the hard segments, while the continuous aqueous phase contains the soft segments. A trifunctional chemical cross-linker for shape memory PU synthesis was utilized to eliminate creep and improve the recovery ratios of the final capsules. We observe an anomalous dependence of the recovery ratio with the amount of programmed strain compared to previous SMPs. We develop quantitative characterization methods and theory to show that when dealing with thin-shell objects, alternative parameters to quantify recovery ratios are needed. We show that while achieving 94–99% area recovery ratios, the linear capsule recovery ratios can be as low as 70%. This quantification method allows us to convert from observed linear aspect ratios in capsules to find out unrecovered area strain and stress. The hollow structure of the capsules grants high internal volume for some applications (e.g., drug delivery), which benefit from much higher loading of active ingredients than polymeric particles. The methods we developed for capsule synthesis and programming could be easily scaled up for larger volume applications.