We show that memory, in the form of underdamped angular dynamics, is a crucial ingredient for the collective properties of self-propelled particles. Using Vicsek-style models with an OrnsteinUhlenbeck process acting on angular velocity, we uncover a rich variety of collective phases not observed in usual overdamped systems, including vortex lattices and active foams. In a model with strictly nematic interactions the smectic arrangement of Vicsek waves giving rise to global polar order is observed. We also provide a calculation of the effective interaction between vortices in the case where a telegraphic noise process is at play, explaining thus the emergence and structure of the vortex lattices observed here and in motility assay experiments.PACS numbers: 05.65.+b, 45.70.Vn, 87.18.Gh Self-propelled particles are nowadays commonly used to study collective motion and more generally "dry" active matter, where the surrounding fluid is neglected. Real world relevant situations include shaken granular particles [1][2][3][4][5], active colloids [6][7][8], bio-filaments displaced by motor proteins [9][10][11]. The trajectories of moving living organisms (from bacteria to large animals such as fish, birds and even human crowds) are also routinely modeled by such particles, see e.g. [12][13][14][15][16][17].Many of these 'active particles' travel at near-constant speed with their dynamics modeled as a persistent random walk with some stochastic component acting directly on their orientation [18]. This noise, which represents external and/or internal perturbations, produces jagged irregular trajectories. Most of the recent results on active matter have been obtained in this context of overdamped dynamics.In many situations, however, the overdamped approximation is not justified. In particular, trajectories can be essentially smooth, as for chemically propelled rods [19,20], birds, some large fish that swim steadily [16], or even biofilaments in motility assays with a high density of molecular motors [10]. Whether underdamped dynamics can make a difference at the level of collective asymptotic properties is largely unknown. Interesting related progress was recently reported for starling flocks [21]. Underdamped "spin" variables are instrumental there for efficient, fast transfer of information through the flock, allowing swift turns in response to threats during which speed is modulated in a well coordinated manner. In the other examples cited above, speed remains nearlyconstant and the persistently turning tracks of fish or microtubules reveal some finite, possibly large, memory of the curvature. In this context an Ornstein-Uhlenbeck (OU) process acting on the angular velocity was shown to be a quantitatively-valid representation [10,16]. The collective motion of self-propelled particles with such underdamped angular dynamics remains largely unknown.In this Letter, we explore minimal models of aligning self-propelled particles with memory similar to that used in [10] to study the emergence of large-scale vortices in mot...