From the perspective of cellular-automaton and by exploiting the interactions between travelling and localized waves the Lorentz Excitable Lattice Gas Automata (LELGA) is proposed and applied to the optimization of Lennard-Jones (LJ) atomic clusters. Our approach exploits the fact that a reaction-diffusion phenomenon with self-localized excitations cells which behave like quasi-particles, could be potentially used to implement dynamical computation. As an example of collision-based computation in reaction-diffusion systems based on cellular automata we provided our experimental results for unbiased optimization of Lennard-Jones (LJ) atomic clusters. We also demonstrated that the proposed method has successfully located the known global minima of LJ clusters for configuration space size of N <= 383.