Continuous Foraging and Information Gathering in a Multi-Agent Team

We are interested in continuous foraging with multi-agent teams, where resources are replenished over time, and the goal is to maximize the rate of foraging. Existing algorithms for continuous foraging and area sweeping typically consider homogeneous agents. We are interested in heterogeneous teams, where agents have radically different capabilities. In particular, we consider two types of agents: a foraging agent that moves in the environment and forages resources, and a reconnaissance agent that gathers information by visiting locations and determining the number of resources available. In this paper, we consider three models of resource replenishment: a Bernoulli model and a Poisson model where resources appear probabilistically, and a stochastic Logistic model where resources increase based on the existing number of resources. We contribute three foraging algorithms that are inspired by existing algorithms, and contribute a novel algorithm for the reconnaissance agent to gather information. We extensively evaluate our algorithms in simulation, showing that our foraging algorithms outperform the existing algorithms. We demonstrate the efficacy of our information gathering algorithm in improving the overall team performance, even without communication between the foraging agents, and with noisy observations.