Switching between collaboration levels in a human-robot target recognition system

This paper presents a set of algorithms that are developed for real-time dynamic switching between collaboration levels in a human-robot target recognition system. The algorithms were developed for a closed-loop controller to maximize system performance, despite deviations in the parameter values. These developments enable smooth real-time adaptation of the combined human-robot system to many possible changes of the environment, human operator, and robot performance. System performance was analyzed in simulations for a variety of target probability distributions. Two hundred independent simulations for each target probability distribution were conducted to calculate algorithm performance for a variety of conditions. Values for human operations were taken from a target recognition experiment dealing with detecting melons for a robotic melon harvester. The numerical analysis results indicated that the developed dynamic switching algorithms resulted in improved system performance that, in some cases, was increased by more than 90%.