The decentralised gathering problem consists in grouping in a compact cluster agents that are initially randomly scattered. We propose a bio-inspired algorithm, the Reaction-Diffusion-Chemotaxis aggregation scheme, to group agents that have limited abilities. The agents and their environment are described with a stochastic model inspired by the aggregation of the {\em Dictyostelium discoideum} cellular slime mold. The environment is an active lattice, whose cells transmit information according to a reaction-diffusion mechanism. The agents are virtual amoebae; they trigger excitations randomly and move by following reaction-diffusion waves. We demonstrate that despite its simplicity, this model exhibits interesting properties of self-organisation and is efficient for gathering agents. Moreover, observations show that the system is robust to various perturbations, such as the presence of obstacles on the lattice or noise in the movements of the agents.