Optimizing Human Attention Allocation in Heterogeneous Semi-Autonomous Robotic Systems Using A Heuristic Computational Framework

Abstract

As robotic systems are increasingly used in military environments, human operators must manage growing multi-robot fleets with limited resources. Optimal human attention allocation is necessary to maintain performance, safety, and effectiveness without increasing operator workload. This study investigates how a single human operator should allocate attention between heterogeneous robots in a semi-autonomous system to maximize performance. A computational framework, structured to reflect the coordination and task-allocation demands of a real-world multi-robot military scenario, modeled a standard soccer field containing 200 collectible resources and 20 scoring targets, with one intake robot and one shooter robot operating simultaneously. Nine workflows were ran through the computational framework to compare human-robot task allocation policies: Fixed Rotation — where autonomous and human control was switched every minute—, Greedy Shooter — where the human always controls the shooter unless the shooter is empty —, Start-Up Model — in which the human controls the intake for the first 30 balls and then switches to shooter for the rest of the game —, and six Split Attention strategies (100% – 0% shooter focus) — in which the human controls the shooter in 20% increments, ranging from 0% of the game to 100% of the game. The computational framework and workflows were run in order to be a basis for future real-life testing, in which estimated parameter values for shooter and intake data, both human and autonomous, were used. No scientifically significant conclusions could be drawn from this computational framework as estimated parameter values were used, and human error was never a calculated factor; however, the framework suggests further investigation into the Split Attention-100% workflow, Greedy Shooter, and Start-up Model, as these performed the best in the context of the computational framework, significantly producing higher scores than their contemporaries. Future work should investigate whether these workflows are actually applicable and accurate in the real world, using real human and real autonomous robot testing.