We consider swarms of luminous myopic robots that run in synchronous Look-Compute-Move cycles. These robots evolve in a finite grid and are disoriented, i.e., they have neither global compass nor a common chirality. In this context, we propose optimal solutions for the perpetual exploration of a finite grid. Precisely, we investigate optimality in terms of the visibility range, number of robots, number of colors. In more detail, under the optimal visibility range one, we give an algorithm which is optimal w.r.t. the number of robots: it uses three robots and three colors. Under visibility two, we design an algorithm that uses five robots and only one color, i.e., robots are oblivious.