Autonomous task partitioning in robot foraging: an approach based on cost estimation

We propose an approach for autonomous task partitioning in swarms of foraging robots. Task partitioning is the process of decomposing tasks into sub-tasks. Task partitioning impacts tasks execution and associated costs. Our approach is characterized by the use of a cost function, mapping the size of sub-tasks to the overall task cost. The robots model the cost function and use the model to select sub-tasks to perform, aiming to minimize costs. Our approach separates the task partitioning process from task-specific actions and it does not require a priori assumptions to be made about the best partitioning strategy to employ. We study a foraging scenario in which object transportation is performed by different robots, each moving objects for a limited distance. The robots autonomously decide the distance traveled on the basis of our approach. The robots use odometry for navigational purposes; we show that task partitioning reduces the impact of odometry errors and improves performance. We validate our approach using simulation-based experiments. We study how the swarm partitions transportation under a number of experimental conditions characterized by different levels of odometry accuracy, size of the environment and the swarm, and total transportation distance. Our approach leads to partitioning solutions that are appropriate for each condition.