Multiple-place swarm foraging with dynamic depots

Teams of robots can be organized to collectively complete complex real-world tasks, for example collective foraging in which robots search for, pick up, and drop off targets in a collection zone. In the previously proposed central-place foraging algorithm (CPFA), foraging performance decreases as swarm size and search areas scale up: more robots produce more inter-robot collisions and larger search areas produce longer travel distances. We propose the multiple-place foraging algorithm with dynamic depots (\(\hbox {MPFA}_{dynamic}\)) to address these problems. Depots are special robots which are initially distributed in the search area and can carry multiple targets. Depots move to the centroids of the positions of local targets recently detected by robots. The spatially distributed design reduces robot transport time and reduces collisions among robots. We simulate robot swarms that mimic foraging ants using the \(\hbox {MPFA}_{dynamic}\) strategy, employing a genetic algorithm to optimize their behavior in the robot simulator ARGoS. Robots using the \(\hbox {MPFA}_{dynamic}\) find and collect targets faster than both the CPFA and the static MPFA. \(\hbox {MPFA}_{dynamic}\) outperforms the static MPFA even when the static depots are optimally placed using global information, and it outperforms the CPFA even when the dynamic depots deliver targets to a central location. Further, the \(\hbox {MPFA}_{dynamic}\) scales up more efficiently, so that the improvement over the CPFA and the static MPFA is even greater in large (50 \(\times \) 50 m) areas. Including simulated error reduces foraging performance across all algorithms, but the MPFA still outperforms the other approaches. Our work demonstrates that dispersed agents that dynamically adapt to local information in their environment provide more flexible and scalable swarms. In addition, we illustrate a path to implement the \(\hbox {MPFA}_{dynamic}\) in the physical robot swarm of the NASA Swarmathon competition.