Lattice gas automata are a powerful tool to model reaction-diffusion processes. However, the evolution rules limit the number of particles that can be present at each lattice site. This restriction is often a strong limitation to modelling several reaction-diffusion phenomena and, also, lead to very noisy numerical simulations. We propose and study new algorithms which allow for an arbitrary number of particle, while keeping the benefits of the cellular automata approach (close to a microscopic level of description, deal correctly with all degrees of freedom, and is numerically exact).