An Approximate Method for Spatial Task Allocation in Partially Observable Environments

Abstract

Multi-robot task allocation has many applications in the real world. Robots often have noisy or local sensor readings, making their workspace partially observable. This paper proposes a partially observable spatial task allocation algorithm, called POSA, that extends the subjective self-absorbed view of E-FWD, a task allocation algorithm for a fully observable environment. POSA uses Partially Observable Monte-Carlo Planning (POMCP) to evaluate the value of the successor belief states. Simulations show that POSA can reach the performance of E-FWD, even though it has partial observability rather than full observability. POSA also has a better convergence rate because it uses Monte-Carlo simulations that estimate the value of suitable locations of search space and does not have to evaluate the value of all parts of the search space.