Collective sampling of environmental features under limited sampling budget

Exploration of an unknown environment is one of the most prominent tasks for multi-robot systems. In this paper, we focus on the specific problem of how a swarm of simulated robots can collectively sample a particular environment feature. We propose an energy-efficient approach for collective sampling, in which we aim to optimize the statistical quality of the collective sample while each robot is restricted in the number of samples it can take. The individual decision to sample or discard a detected item is performed using a voting process, in which robots vote to converge to the collective sample that reflects best the inter-sample distances. These distances are exchanged in the local neighbourhood of the robot. We validate our approach using physics-based simulations in a 2D environment. Our results show that the proposed approach succeeds in maximizing the spatial coverage of the collective sample, while minimizing the number of taken samples.