Inference Model for Heterogeneous Robot Team Configuration based on Reinforcement Learning

In many practical robotics problems, knowledge of the team configuration and capabilities is crucial in coordination of multiple heterogeneous robots. In a challenging environment with costly, sporadic, or absent communication, inferencing based on observed spatio-temporal state transitions is necessary for learning and reasoning. In this paper, we present a general purpose inference engine that takes sparse observations of state transitions made during multi-robot team execution of a foraging task as input and dynamically inferences the team configuration through a rational decision-making process using Reinforcement Learning (RL). We demonstrate the operation and scalability of this approach in simulations using various size multi-robot foraging tasks. The method is robust to dynamic changes in team composition during execution.